U.S. patent number 10,929,641 [Application Number 16/057,710] was granted by the patent office on 2021-02-23 for smart microscope system for radiation biodosimetry.
This patent grant is currently assigned to CytoGnomix Inc.. The grantee listed for this patent is Yanxin Li, Jin Liu, Peter Keith Rogan. Invention is credited to Yanxin Li, Jin Liu, Peter Keith Rogan.
![](/patent/grant/10929641/US10929641-20210223-D00000.png)
![](/patent/grant/10929641/US10929641-20210223-D00001.png)
![](/patent/grant/10929641/US10929641-20210223-D00002.png)
![](/patent/grant/10929641/US10929641-20210223-D00003.png)
![](/patent/grant/10929641/US10929641-20210223-D00004.png)
![](/patent/grant/10929641/US10929641-20210223-D00005.png)
![](/patent/grant/10929641/US10929641-20210223-D00006.png)
![](/patent/grant/10929641/US10929641-20210223-D00007.png)
![](/patent/grant/10929641/US10929641-20210223-D00008.png)
![](/patent/grant/10929641/US10929641-20210223-D00009.png)
![](/patent/grant/10929641/US10929641-20210223-D00010.png)
United States Patent |
10,929,641 |
Rogan , et al. |
February 23, 2021 |
Smart microscope system for radiation biodosimetry
Abstract
An automated microscope system is described that detects
dicentric chromosomes (DCs) in metaphase cells arising from
exposure to ionizing radiation. The radiation dose depends on the
accuracy of DC detection. Accuracy is increased using image
segmentation methods are used to rank high quality cytogenetic
images and eliminate suboptimal metaphase cell data in a sample
based on novel quality measures. When a sufficient number of high
quality images are detected, the microscope system is directed to
terminate metaphase image collection for a sample. The microscope
system integrates image selection procedures that control an
automated digitally controlled microscope with the analysis of
acquired metaphase cell images to accurately determine radiation
dose. Early termination of image acquisition reduces sample
processing time without compromising accuracy. This approach
constitutes a reliable and scalable solution that will be essential
for analysis of large numbers of potentially exposed
individuals.
Inventors: |
Rogan; Peter Keith (London,
CA), Li; Yanxin (Kitchener, CA), Liu;
Jin (London, CA) |
Applicant: |
Name |
City |
State |
Country |
Type |
Rogan; Peter Keith
Li; Yanxin
Liu; Jin |
London
Kitchener
London |
N/A
N/A
N/A |
CA
CA
CA |
|
|
Assignee: |
CytoGnomix Inc. (London,
CA)
|
Family
ID: |
1000005378511 |
Appl.
No.: |
16/057,710 |
Filed: |
August 7, 2018 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20200050831 A1 |
Feb 13, 2020 |
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06K
9/00134 (20130101); G06T 5/50 (20130101); G06K
9/0014 (20130101); G01T 1/02 (20130101); G06K
9/00147 (20130101); G06T 5/001 (20130101); G06T
2207/10056 (20130101); G06T 2207/20224 (20130101) |
Current International
Class: |
G06K
9/00 (20060101); G06T 5/00 (20060101); G01T
1/02 (20060101); G06T 5/50 (20060101) |
Field of
Search: |
;382/128,133 |
References Cited
[Referenced By]
U.S. Patent Documents
Other References
Rogan PK et al. (2016) Radiation Dose Estimation by Automated
Cytogenetic Biodosimetry, Radiation Protection Dosimetry 172,
207-217. cited by applicant .
MATLAB (R2014a), Mathworks, Natick, MA. cited by applicant .
Rieder C and Palazzo R (1992) Colcemid and the mitotic cycle,
Journal of cell science 102 ( Pt 3), 387-392. cited by applicant
.
Sethakulvichai, W., et al (2012) Estimation of band level
resolutions of human chromosome images, International Joint Conf.
Comp Science and Software Engineering, pp. 276-282. cited by
applicant .
Carothers, A., and Piper, J. (1994) Computer-aided classification
of human chromosomes: a review, Statistics and Computing 4,
161-171. cited by applicant .
International Standing Committee on Human Cytogenetic Nomenclature,
Shaffer et al.. (2013) ISCN 2013: An International System for Human
Cytogenetic Nomenclature (2013), Karger. cited by applicant .
Schunck C et al. (2004) New developments in automated cytogenetic
imaging: unattended scoring of dicentric chromosomes, micronuclei .
. . Cytogenet. Genome Res. 104, 383-389. cited by applicant .
Blakely W et al. (2005) Early-response biological
dosimetry--recommended countermeasure enhancements for
mass-casualty radiological incidents . . . Health physics 89,
494-504. cited by applicant .
Wilkins R et al. (2008) Interlaboratory comparison of the dicentric
chromosome assay for radiation biodosimetry in mass casualty
events. Radiation Research 169, 551-560. cited by applicant .
Bauchinger M (1984)Cytogenetic effects in human lymphocytes as a
dosimetry system. In: Eisert WS, Mendelsohn ML ed. Biological
dosimetry: . . . Springer-Verlag; 15-24. cited by applicant .
Lloyd DC et al. (1986) Chromosome aberrations induced in human
lymphocytes by in vitro acute X and Gamma radiation. Rad. Prot.
Biodosimetry. 15:83-88. cited by applicant .
International Atomic Energy Agency (IAEA). Cytogenetic analysis for
radiation dose assessment. Technical Report Series No. 405, Vienna
(2001). cited by applicant .
International Atomic Energy Agency (IAEA). Cytogenetic Dosimetry:
Applications in Preparedness for and Response to Radiation
Emergencies, Vienna (2011). cited by applicant .
Ainsbury EA et al. (2009) Interlaboratory variation in scoring
dicentric chromosomes in a case of partial-body x-ray exposure:
implications for . . . Radiat. Res. 172: 746-752. cited by
applicant .
Lloyd DC et al. (2000) The role of cytogenetics in early triage of
radiation casualties. Appl. Radiat. Isot. 52:1107-1112. cited by
applicant .
Flegal F et al. (2010) Quickscan dicentric chromosome analysis for
radiation biodosimetry. Health Phys. 98: 276-281. cited by
applicant .
Vaurijoux A et al. (2009) Strategy for population triage based on
dicentric analysis. Radiat Res 171:541-548. cited by applicant
.
Vaurijoux A et al. (2015) Automatic Dicentric Scoring a Real Option
to Be Used in Biological Dosimetry. Radiation Emergency Medicine.
4:16-21. cited by applicant .
Gruel G et al. (2013) Biological Dosimetry by Automated Dicentric
Scoring in a Simulated Emergency. Radiation Res. 179: 557-569.
cited by applicant .
Wang Z et al. (2004) Image quality assessment: from error
visibility to structural similarity, IEEE Transactions on Image
Processing 13, 600-612. cited by applicant .
Nill NB and Bouzas B (1992) Objective image quality measure derived
from digital image power spectra, OPTICE 31, 813-825. cited by
applicant .
Narwaria M and Lin W (2010) Objective Image Quality Assessment
Based on Support Vector Regression, IEEE Transactions on Neural
Networks 21, 515-519. cited by applicant .
Li Y et al. (2016) Automated discrimination of dicentric and
monocentric chromosomes by machine learning-based image processing,
Microscopy Research and Technique 79, 383-402. cited by applicant
.
Arachchige AS et al (2010) An image processing algorithm for
accurate extraction of the centerline from human metaphase . . .
IEEE Int. Conf Image Processing pp. 3613-3616. cited by applicant
.
Arachchige AS et al (2012) Intensity integrated Laplacian algorithm
for human metaphase chromosome centromere detection, IEEE Can Conf.
Electrical & Computer Engineering pp. 1-4. cited by applicant
.
Arachchige AS et al.(2013) Intensity integrated Laplacian-based
thickness measurement for detecting human metaphase chromosome
centromere . . . IEEE Trans.Biomed Eng.60: 2005-13. cited by
applicant .
Subasinghe A et al. (2016) Centromere Detection of Human Metaphase
Chromosome Images using a Candidate Based Method, F1000Research 5,
1565, 2016. cited by applicant.
|
Primary Examiner: Sherali; Ishrat I
Attorney, Agent or Firm: Ingenium Patents LLC
Claims
We claim:
1. A method of estimation of a radiation exposure by biodosimetry
in a sample of cells, said sample of cells prepared for cytogenetic
analysis from a single individual, said method performed using an
automated digitally controlled microscope system, said method
comprising: (i) acquiring images of cells sequentially, wherein
said images of cells contain metaphase chromosomes, and further
wherein said images of cells are acquired by using said automated
digitally controlled microscope system, the microscope system
having a microscope with a computer-controlled digital camera, (ii)
digitally analyzing objects in each image of said images of cells
to determine a property or properties of segmented objects therein,
said property or properties including object count, length, width,
contour finite difference. and centromere density, (iii) selecting
or rejecting said each image of said images based on said property
or properties determined in the preceding step, thereby creating a
set of selected digital images, (iv) directing the microscope
system to discontinue the acquisition of images of step (i) after a
sufficient number of images have been captured to determine a
radiation dose, thereby generating a set of images containing
metaphase chromosomes, (v) creating a set of likely dicentric
chromosomes by classifying likely dicentric chromosomes in the set
of selected digital images from step (iii), and determining a count
of the likely dicentric chromosomes in the set of selected digital
images. (vi) determining which chromosomes of the set of likely
dicentric chromosomes from step (v) are not true dicentric
chromosomes using segmentation procedures that discriminate true
positive dicentric chromosomes from other objects, thereby
identifying false positive dicentric chromosomes and determining a
count of false positive dicentric chromosomes in the set of
selected digital images. (vii) eliminating the set of selected
digital images of false positive dicentric chromosomes from the set
of likely dicentric chromosomes, (viii) determining a numerical
count of false positive dicentric chromosomes and determining a
count of the dicentric chromosomes in each digital image by
subtracting the number of false positive dicentric chromosomes from
a total number of the likely dicentric chromosomes in each image,
(ix) determining a dose response for the sample, said dose response
being an average dicentric chromosome frequency over all images
from the sample, by summing the total number of corrected dicentric
chromosomes in said set of images containing metaphase chromosomes
from the sample and dividing by the number of images in said set of
images containing metaphase chromosomes, (x) computing the
radiation exposure using a previously determined dose response
related calibration curve that is related to the dose response by
the quadratic equation, Y=aX.sup.2+bX+c wherein a, b, and c are
coefficients of the curve, and wherein X denotes dose response and
Y denotes radiation exposure, (xi) sending a signal to the
digitally controlled microscope system indicating that the process
of collecting images from a sample has been completed, and
terminating the collection of new image data for that sample.
2. The method of estimation of radiation exposure by biodosimetry
of claim 1, said classification of a predicted dicentric
chromosome, c*, in a metaphase cell digital image as a false
positive dicentric chromosome, where {c.sub.1, . . . , c.sub.N}
denotes the set of N chromosomes within the image, said predicted
dicentric chromosome fulfilling any one or more of the following
conditions, which are performed either independently or in
combination: (i) classifying a predicted dicentric chromosome, c*,
as a false positive dicentric chromosome, if the pixel area, A(c),
occupied by the chromosome, is related to the areas of all other
chromosomes in the same metaphase cell according to:
A(c*)/median({A(c.sub.1), . . . , A(c.sub.N)})<0.74 (ii)
classifying a predicted dicentric chromosome, c*, as a false
positive dicentric chromosome, in which W.sub.mean(c) denotes the
mean value of the width profile of chromosome c, and
W.sub.mean(c*)/median({W.sub.mean(c.sub.1), . . . ,
W.sub.mean(c.sub.N) })<0.80, (iii) classifying a predicted
dicentric chromosome, c*, as a false positive dicentric chromosome,
in which W.sub.med(c) denotes the median value of the width profile
of chromosome c, and W.sub.med(c*)/median({W.sub.mean(c.sub.1), . .
. , W.sub.mean(c.sub.N) })<0.77, (iv) classifying a predicted
dicentric chromosome, c*, as a false positive dicentric chromosome,
in which W.sub.max(C) denotes the maximum value of the width
profile of chromosome c, and W.sub.max(c*)/median({W.sub.max(c1), .
. . , W.sub.max(c.sub.N) })<0.83, (v) classifying a predicted
dicentric chromosome, c*, as a false positive dicentric chromosome,
in which W.sub.cent(c) (denote the width of chromosome c at the
top-ranked centromere candidate, and
W.sub.cent(c*)median({W.sub.cent(c.sub.1), . . . ,
W.sub.cent(c.sub.n)})<0.72, (vi) classifying a predicted
dicentric chromosome, c*, as a false positive dicentric chromosome,
in which S(c) denotes the pair of side lengths of the minimum
bounding rectangle enclosing the contour of chromosome c, and
1-min(S(c*))/max(S(c*))<0.28, (vii) classifying a predicted
dicentric chromosome, c*, as a false positive dicentric chromosome,
in which L(c) denotes the pair of arc lengths of contour halves
produced by partitioning the contour of chromosome c at its
centerline endpoints, and min(L(c*))/max(L(c*))<0.51, (viii)
classifying a predicted dicentric chromosome, c*, as a false
positive dicentric chromosome, in which L.sub.c(c) denotes the pair
of arc lengths of the contour regions of chromosome c that run
between the traceline endpoints of its top 2 centromere candidates,
and min(L.sub.c(c*))/max(L.sub.c(c*))<0.42.
3. The method of estimation of radiation exposure by biodosimetry
of claim 1, said digital analysis of images of cells from the same
sample, with each image containing chromosomes from a cell in
metaphase, the sample comprising M images, {I.sub.1, . . . ,
I.sub.M}, where {c.sub.1,. . . , c.sub.N} denote the set of N
chromosomes within image I*, and SD denotes the standard deviation
function, and T denotes the threshold standard deviation value that
identifies outlier images, said method, after applying filters,
that either individually or combination, determines whether an
image shall be removed from the sample, the digital filters
comprising the following steps either individually or in
combination: (i) applying the Length-width ratio filter (LW) which
defines the average length-width ratio of chromosomes in an image.
For a given chromosome c in a given image I containing N
chromosomes, L(c,I) denotes the arc length of the centerline of c,
and W.sub.mean(c,I) denotes the mean value of the width profile of
c. MW(I) is defined as the mean{L(c.sub.1,I)/W.sub.mean(c.sub.1,
I), . . . ,L(C.sub.N,I)/W.sub.mean(C.sub.N,I)} length-width ratio.
I* is removed if MW(I*)>mean{MW(I1), . . . ,
MW(I.sub.M)}+T.times.SD{MW(I.sub.1), . . . , MW(I.sub.M)}, (ii)
applying the Centromere candidate density filter (CD) which counts
occurrences of centromere candidates in images of chromosomes. For
a given chromosome c in a given image I containing N chromosomes,
L(c,I) denotes the arc length of the centerline of c, and
N.sub.cent(c,I) denotes the number of centromere candidates of c.
CD(I) is defined as the mean{N.sub.cent(c.sub.1,I)/L(c.sub.1,I), .
. . , N.sub.cent(c.sub.N,I)/L(c.sub.N,I)}. I* is removed if
CD(I*)>mean{CD(I.sub.1), . . . ,
CD(I.sub.M)}+T.times.SD{CD(I.sub.1), . . . , CD(I.sub.M)}, (iii)
applying Contour finite difference filter (FD) which represents
contour smoothness of chromosomes in an image. For a given
chromosome c in a given image I containing N chromosomes,
WP.sub.D(c,I) denotes the set of first differences of the
normalized width profile of c (range normalized to interval [0,1]).
WD(I) is defined as the mean{mean{abs{WP.sub.D(c.sub.1,I)}}, . . .
, mean{abs{WP.sub.D(c.sub.N,I)}}}. I* is removed if
WD(I*)<mean{WD(I.sub.1), . . . ,
WD(I.sub.M)}-T.times.SD{WD(I.sub.1), . . . , WD(I.sub.M)}, (iv)
applying the Total object count (Obj Count) filter, which defines
the number of all objects, O, including chromosomes and
non-chromosomal objects detected in an image. I* is removed if
O<40or O>60, (v) applying the Segmented object count
(SegObjCount) filter, which defines the number of objects processed
by the gradient vector flow algorithm, O.sub.GVF, in an image. I*
is removed if O.sub.GVF<35 or O.sub.GVF>50, (vi) applying the
Classified object ratio (ClassifiedRatio) filter, which defines the
ratio of objects recognized as chromosomes, N, to the number of
segmented objects, O.sub.GVF. The stringency of this filter may be
configured by adjusting the threshold of the acceptable minimum
ratio to be either permissive (lower) or strict (higher), so that
lower. I* is removed N/O.sub.GVF<0.6 (permissive) or 0.7
(strict).
4. The method of estimation of radiation exposure by biodosimetry
of claim 3, wherein said digital analysis of images of metaphase
cells from the same sample, which determines a composite filter
score computed from each of the filter values defined as the
elements (i), (ii), (iii), (iv), (v), and (vi) of claim 3, said
method further comprising the following steps: (i) combining one or
more Z-scores of each of the filters for an image relative to the
population of M images in a sample using the following linear
expression: Composite Filter
Score=w(LW)*z(LW)+w(CD)*z(CD)-w(FD)*z(FD)+w(ObjCount)*|z(ObjCount)|+w(Seg-
ObCount)*|z(SegObjCount)|-w(ClassifiedRatio)*z(Classified Ratio)
where each of the filters, LW, CD, FD, Obj Count, SegObj Count, and
ClassifiedRatio, contains a positive free parameter, weight (w) to
adjust its contribution to the total score, and w is determined by
evaluating and selecting values that minimize the deviation from
known physical dose in a dose calibration curve, (ii) ranking each
of the images in a sample based on the score, such that the highest
scores are obtained for images exhibiting either incomplete,
multiple cells or severe sister chromatid separation, or images
that the automated dicentric detection algorithm does not process
accurately (iii) and removing the images with the largest combined
Z-values, which have the largest composite filter scores from the
sample.
5. The method of estimation of radiation exposure levels by
biodosimetry of claim 2, wherein said digital analysis of images of
cells from the same sample, with each image containing chromosomes
from a cell in metaphase, the sample comprising M images, {I.sub.1,
. . . , I.sub.M}, where {c.sub.1, . . . , c.sub.N} denote the set
of N chromosomes within image I*, and SD denotes the standard
deviation function, and T denotes the threshold standard deviation
value that identifies outlier images, said method, after applying
filters, that either individually or combination, determines
whether an image shall be removed from the sample, the digital
filters comprising the following steps either individually or in
combination: (i) applying the Length-width ratio filter (LW) which
defines the average length-width ratio of chromosomes in an image.
For a given chromosome c in a given image I containing N
chromosomes, L(c,I) denotes the arc length of the centerline of c,
and W.sub.mean(c,I) denotes the mean value of the width profile of
c. MW(I) is defined as the mean{L(c.sub.1,I)/W.sub.mean(C.sub.1,I),
. . . , L(C.sub.N,I)W.sub.mean(C.sub.N,I)} length-width ratio. I*
is removed if MW(I*)>mean {MW(I1), . . . ,
MW(I.sub.M)}+T.times.SD {MW(I.sub.1), . . . , MW(I.sub.M)}, (ii)
applying the Centromere candidate density filter (CD) which counts
occurrences of centromere candidates in images of chromosomes. For
a given chromosome c in a given image I containing N chromosomes,
L(c,I) denotes the arc length of the centerline of c, and
N.sub.cent(c,I) denotes the number of centromere candidates of c.
CD(I) is defined as the mean{N.sub.cent(c,I)/L(c.sub.1,I), . . . ,
N.sub.cent(c.sub.N,I)/L(c.sub.N,I)}. I* is removed if
CD(I*)>mean{CD(I.sub.1), . . . , CD(I.sub.M)}+T.times.SD
{CD(I.sub.1), . . . , CD(I.sub.M)}, (iii) applying Contour finite
difference filter (FD) which represents contour smoothness of
chromosomes in an image. For a given chromosome c in a given image
I containing N chromosomes, WP.sub.D(c,I) denotes the set of first
differences of the normalized width profile of c (range normalized
to interval [0,1]). WD(I) is defined as the
mean{mean{abs{WP.sub.D(c.sub.1,I)}} , . . . ,
mean{abs{WP.sub.D(c.sub.N,I)}}}. I* is removed if
WD(I*)<mean{WD(I.sub.1), . . . , WD(I.sub.M)}-T.times.SD
{WD(I.sub.1), . . . , WD(I.sub.M)}, (iv) applying the Total object
count (Obj Count) filter, which defines the number of all objects,
O, including chromosomes and non-chromosomal objects detected in an
image. I* is removed if O<40 or O>60, (v) applying the
Segmented object count (SegObjCount) filter, which defines the
number of objects processed by the gradient vector flow algorithm,
O.sub.GVF, in an image. I* is removed if O.sub.GVF<35 or
O.sub.GVF>50, (vi) applying the Classified object ratio
(ClassifiedRatio) filter, which defines the ratio of objects
recognized as chromosomes, N, to the number of segmented objects,
O.sub.GVF. The stringency of this filter may be configured by
adjusting the threshold of the acceptable minimum ratio to be
either permissive (lower) or strict (higher), so that lower. I* is
removed N/O.sub.GVF<0.6 (permissive) or 0.7 (strict).
6. The method of improving accuracy of estimation of radiation
exposure levels by biodosimetry of claim 2, said method further
comprising the following steps: (i) combining one or more Z-scores
of each of the filters for an image relative to the population of M
images in a sample using the following linear expression: Composite
Filter
Score=w(LW)*z(LW)+w(CD)*z(CD)-w(FD)*z(FD)+w(ObjCount)*|z(ObjCount)|+w(Seg-
ObCount)*|z(SegObjCount)|-w(ClassifiedRatio)*z(Classified Ratio)
where each of the filters, LW, CD, FD, Obj Count, SegObj Count, and
ClassifiedRatio, contains a positive free parameter, weight (w) to
adjust its contribution to the total score, and w is determined by
evaluating and selecting values that minimize the deviation from
known physical dose in a dose calibration curve, (ii) ranking each
of the images in a sample based on the score, such that the highest
scores are obtained for images exhibiting either incomplete,
multiple cells or severe sister chromatid separation, or images
that the automated dicentric detection algorithm does not process
accurately (iii) and removing the images with the largest combined
Z-values, which have the largest composite filter scores from the
sample.
7. The method of estimation of radiation exposure levels by
biodosimetry of claim 2, said method removing false positive
dicentric chromosomes from images of metaphase cells of claim 2,
and selecting metaphase images by digital analysis of images of
cells from the same sample, with each image containing chromosomes
from a cell in metaphase, the sample comprising M images, {I.sub.1,
. . . , I.sub.M}, where {c.sub.1, . . . , c.sub.N} denote the set
of N chromosomes within image I*, and SD denotes the standard
deviation function, and T denotes the threshold standard deviation
value that identifies outlier images, said method, after applying
filters, that either individually or combination, determines
whether an image shall be removed from the sample, the digital
filters comprising the following steps either individually or in
combination: (i) applying the Length-width ratio filter (LW) which
defines the average length-width ratio of chromosomes in an image.
For a given chromosome c in a given image I containing N
chromosomes, L(c,I) denotes the arc length of the centerline of c,
and W.sub.mean(c,I) denotes the mean value of the width profile of
c. MW(I) is defined as the mean {L(c.sub.1I)/W.sub.mean(c.sub.1,I),
. . . , L(c.sub.N,I)} length-width ratio. I* is removed if
MW(I*)>mean{MW(I1), . . . , MW(I.sub.M)}+T.times.SD{MW(I.sub.1),
. . . , MW(I.sub.M)}, (ii) applying the Centromere candidate
density filter (CD) which counts occurrences of centromere
candidates in images of chromosomes. For a given chromosome c in a
given image I containing N chromosomes, L(c,I) denotes the arc
length of the centerline of c, and N.sub.cent(c,I) denotes the
number of centromere candidates of c. CD(I) is defined as the
mean{N.sub.cent(c.sub.1,I)/L(c.sub.1,I), . . . ,
N.sub.cent(c.sub.NI)}. I* is removed if CD(I*)>mean{CD(I.sub.1),
. . . , CD(I.sub.M)}+T.times.SD {CD(I.sub.M)}, (iii) applying
Contour finite difference filter (FD) which represents contour
smoothness of chromosomes in an image. For a given chromosome c in
a given image I containing N chromosomes, WP.sub.D(c,I) denotes the
set of first differences of the normalized width profile of c
(range normalized to interval [0,1]). WD(I) is defined as the
mean{mean{abs{WP.sub.D(c.sub.1,I)}}, . . . ,
mean{abs{WP.sub.D(c.sub.N,I)}}}. I* is removed if
WD(I*)<mean{WD(I.sub.1), . . . , WD(I.sub.M)}-T.times.SD
{WD(I.sub.1), . . . , WD(I.sub.M)}, (iv) applying the Total object
count (ObjCount) filter, which defines the number of all objects,
O, including chromosomes and non-chromosomal objects detected in an
image. I* is removed if O<40 or O>60, (v) applying the
Segmented object count (SegObjCount) filter, which defines the
number of objects processed by the gradient vector flow algorithm,
O.sub.GVF, in an image. I* is removed if O.sub.GVF<35 or
O.sub.GVF>50. (vi) applying the Classified object ratio
(ClassifiedRatio) filter, which defines the ratio of objects
recognized as chromosomes, N, to the number of segmented objects,
O.sub.GVF. The stringency of this filter may be configured by
adjusting the threshold of the acceptable minimum ratio to be
either permissive (lower) or strict (higher), so that lower. I* is
removed N/O.sub.GVF<0.6 (permissive) or 0.7 (strict).
8. The method of estimation of radiation exposure levels by
biodosimetry in a sample from an individual of claim 2, said method
removing false positive dicentric chromosomes from images of
metaphase cells of claim 2, and selecting metaphase images 4 by
digital analysis of images of metaphase cells from the same sample,
which determines a composite filter score computed from each of the
filter values, said method comprising the following steps: (i)
combining one or more Z-scores of each of the filters for an image
relative to the population of M images in a sample using the
following linear expression: Composite Filter
Score=w(LW)*z(LW)+w(CD)*z(CD)-w(FD)*z(FD)+w(ObjCount)*|z(ObjCount)|+w(Seg-
ObCount)*|z(SegObjCount)|-w(ClassifiedRatio)*z(Classified Ratio)
where each of the filters, LW, CD, FD, Obj Count, SegObj Count, and
ClassifiedRatio, contains a positive free parameter, weight (w) to
adjust its contribution to the total score, and w is determined by
evaluating and selecting values that minimize the deviation from
known physical dose in a dose calibration curve, (ii) ranking each
of the images in a sample based on the score, such that the highest
scores are obtained for images exhibiting either incomplete,
multiple cells or severe sister chromatid separation, or images
that the automated dicentric detection algorithm does not process
accurately (iii) and removing the images with the largest combined
Z-values, which have the largest composite filter scores from the
sample.
9. The method of estimation of radiation exposure levels by
biodosimetry in a sample from an individual of claim 2, said method
wherein false positive dicentric chromosomes from images of
metaphase cells of claim 2 are removed at said eliminating the set
of selected digital images of false positive dicentric chromosomes
step of claim 1.
10. The method of estimation of radiation exposure levels by
biodosimetry in a sample from an individual of claim 3, said method
selecting metaphase images according to claim 3.
11. The method of estimation of radiation exposure levels by
biodosimetry in a sample from an individual of claim 4, said method
selecting metaphase images according to claim 4, said method
selecting metaphase images according to claim 4, further comprising
any or all of the following steps: (i) reducing the size of a
confidence interval of the estimated exposure, wherein the size of
the reduced confidence interval is less than the interval computed
from the unselected set of metaphase images, (ii) reducing the dose
estimation error to within 0.5 Gy of the corresponding physical
radiation dose, (iii) demonstrating that dicentric chromosome
counts among a set of selected metaphase images from the same
sample are Poisson distributed thereby improving the quality of
image data of said selected metaphase images.
12. The method of estimation of radiation exposure levels by
biodosimetry of claim 1, wherein the automatic selection of digital
images obtained from metaphase cells from a sample isolated from an
individual is performed by ranking images with a score computed
from the known lengths of chromosomes, which are proportionate to
the known base-pair counts of each complete chromosome, whereby the
quality of a metaphase cell image is determined by comparing
distribution of observed chromosome object lengths with the
expected distribution of lengths obtained from relative known
base-pair counts of chromosome in the reference human genome
sequence, as follows: (i) the individual chromosome lengths in each
image are approximated according to their corresponding chromosome
areas in pixels, (ii) a fractional area of each chromosome relative
to the total area of all chromosomes is determined, (iii) the
chromosomes are binned according to base-pair lengths into
categories corresponding to grouping defined by the International
System of Cytogenetic Nomenclature, namely (1) groups A and B,
which contain >2.9% of the DNA, (2) group C, which contains
between 2 and 2.9% of DNA, and (3) groups D, E, F, and G, which
contain <2% of the DNA (4) X chromosome, which contains
approximately 2.9% of the DNA, and (5) Y chromosome which contains
approximately 2% of the DNA, of the total base-pairs in a complete
chromosome set, (iv) the thresholds in (iii) are compared to the
fractional area of each chromosome in the metaphase image,
accounting for the correct length of the sex chromosomes by
reference to the known sex of the individual from whom the sample
was obtained, by categorizing the result for each of the three bins
in an image as a 3-element vector, and calculating a Euclidean
distance from the vector to an idealized vector based on the
reference human chromosome lengths, (v) sorting and ranking these
Euclidean distances for all images in a sample, (vi) and
eliminating images from a sample with the largest Euclidean
distances, which exhibit the lowest similarity to the chromosome
length distributions in a normal karyotype.
13. The method of estimation of radiation exposure levels by
biodosimetry in a sample of an individual of claim 3, said method
further comprising at least one of steps (i)-(iii) below (i)
reducing the size of a confidence interval of the estimated
exposure wherein the size of the reduced confidence interval is
less than the interval computed from the unselected set of
metaphase images, (ii) reducing the dose estimation error to within
0.5 Gy of the corresponding physical radiation dose, (iii)
demonstrating that dicentric chromosome counts among a set of
selected metaphase images from the same sample are Poisson
distributed, thereby improving the quality of image data of said
selected metaphase images of claim 3.
14. The method of estimation of radiation exposure levels by
biodosimetry in a sample of an individual of claim 2, said method
further comprising at least one of steps (i)-(iii) below, (i)
reducing the size of a confidence interval of the estimated
exposure, wherein the size of the reduced confidence interval is
less than the interval computed from the unselected set of
metaphase images, (ii) reducing the dose estimation error to within
0.5 Gy of the corresponding physical radiation dose, (iii)
demonstrating that dicentric chromosome counts among a set of
selected metaphase images from the same sample are Poisson
distributed thereby improving the quality of image data of said
selected metaphase images.
15. The method of estimation of radiation exposure levels by
biodosimetry of claim 12, further comprising: determined by: (i)
determining an observed distribution of dicentric chromosomes in
all of the cell images in the sample according to the number of
cells containing i dicentric chromosomes, where i=0 or an integer
>0, (ii) estimating an expected distribution of dicentric
chromosomes from a Poisson distribution, with the .lamda. parameter
of the distribution set to the average number of dicentric
chromosomes per cell in all of the cell images in the sample, (iii)
computing a Pearson Chi-squared goodness of fit statistic based on
the observed and expected dicentric chromosome distributions for
i-1 degrees of freedom and .alpha.=0.01, (iv) performing steps (i),
(ii), and (iii) for the set of images in the sample after removal
of the low quality images, (v) determining if the sample null
hypothesis that the dicentric chromosomes in the sample follow a
Poisson distribution is rejected for the complete set of images and
accepted for the sample wherein low quality images have been
removed.
16. The method of estimation of radiation exposure levels by
biodosimetry of claim 12, said method further comprising: (i)
selection of a set of samples of known radiation doses, each
consisting of a set of metaphase cell images, (iii) assignment of a
support vector machine sigma value for dicentric chromosome
detection, (iv) assignment of a maximum number of images to be
ranked, (iv) assignment of a range of parameter values spanning the
search space of all possible image selection models that are
evaluated and compared to determine the accuracy of each
combination of parameters, (v) evaluation of one or more parameter
combinations either by selecting the model with the highest
p-values of Poisson fit of dicentric chromosome distribution
(p>0.05) for all samples in the set, or by selecting an optimal
dose calibration curve from the sample set in (i) by minimizing the
residual deviations from the known radiation dose, or by performing
a leave-one cross-validation of the estimated dose for each of the
samples in (i), (vi) presents an optimal automated selection models
found during the search sorted according to the overall accuracy of
dose estimation determined from the root mean squared sum of
differences between the estimated and physical radiation doses over
all samples in the set.
Description
BACKGROUND
The analysis of microscopy images of cells is the basis of several
types of analysis of the effects of damage by ionizing radiation.
The gold standard radiation biodosimetry method, the dicentric
chromosome assay (DCA), involves measuring the frequency of
aberrant dicentric chromosomes in a patient sample. While some
aspects of the assay have been successfully automated and
streamlined, its overall throughput remains limited by the
labour-intensive manual dicentric (DC) scoring step, potentially
affecting timely estimation of radiation exposures of multiple
affected individuals, for example, in a radiation accident or a
mass casualty event (Blakely et al. 2005; Wilkins et al. 2008).
Biodosimetry is a useful tool for assessing the dose received by an
individual when no reliable physical dosimetry is available.
Traditionally, the dicentric chromosome assay is the method of
choice for recent acute exposures to ionizing radiation. This
cytogenetic method is based on measuring the frequency of dicentric
chromosomes (DCs) in metaphase cells and converting this frequency
to dose using in vitro generated calibration curves (Bauchinger et
al. 1984; Lloyd et al. 1986; IAEA 2001). Classical, microscope
analysis of DCs is robust, allowing the estimation of doses in the
range of 0.1 to 5 Gy. For dose estimates in the low end of this
range, however, 1000 cells are typically scored (IAEA 2011) making
this method time consuming and only feasible for small numbers of
exposures. This manual approach lacks adequate throughput for a
mass casualty event to estimate the radiation exposures needed to
triage for diagnosis and treatment.
Other cytogenetic assays and systems have been described for
measuring absorbed radiation. These systems are distinct from the
DCA and suffer from several disadvantages. DCs are among the most
stable biological markers of radiation exposure and can be detected
up to 3 months post exposure. The micronucleus assay, by contrast,
can be performed up to 7 days after exposure. The H2AX assay, which
measures DNA damage, can be used up to 72 hr after radiation. Also,
the DCA can be performed using Giemsa stained chromosomes, which
contrasts with other assays requiring fluorescence in situ
hybridization to identify chromosomes or elements of chromosomes.
The DCA is considerably faster, less expensive, and involves less
complex laboratory procedures than other cytogenetic techniques,
since fluorescent reagents and wash steps following their
application are not required. Fluorescent techniques based on
cytogenetic microscopy include identification chromosome
rearrangements based on translocations with chromosome painting
probes, or to mark chromosomes with centromere and telomere probes.
The RABiT system automates fluorescent-based biodosimetry assays
which do not depend on metaphase chromosome image analysis (U.S.
Pat. No. 7,787,681B2, U.S. Pat. No. 7,822,249B2, U.S. Pat. No.
7,826,977B2, U.S. Pat. No. 7,898,673B2). The system can either
count H2AX protein foci detected by fluorescently labeled
antibodies or can use fluorescent probes or stains to identify and
count binucleated micronuclei in microscopy images. While these
steps are automated, RABiT does not however determine when
sufficient numbers of dicentric chromosomes or cells of adequate
quality have been identified, nor does it determine the level of
exposure in Gy based on a calibration curve. The RABiT microscope
system does not automate manage sample acquisition for radiation
biodosimetry in the same manner that the instant invention performs
these tasks.
In response to the pressing demand to increase throughput in
cytogenetic biodosimetry, capture of metaphase images and
interpretation of DCs have been partially automated, with a
concomitant reduction in the numbers of cell analyzed. Software
(eg. MSearch, DCScore [Metasystems]) has automated the scanning of
microscope slides to locate metaphase cells and assisted review of
DCs for triage biodosimetry (Schunck et al. 2004). This software
has also facilitated inter-laboratory collaboration and the
assessment of partial-body exposures (Ainsbury et al. 2009). The
adoption of triage scoring of 50 carefully selected cells greatly
increases the throughput, while maintaining the ability to identify
exposures of over 1 Gy (Lloyd et al. 2000), and reducing the time
required by more than a factor of 5 (Flegal et al. 2010).
More recently, automated image analysis software that can identify
DCs (DCScore.TM., Metasystems) has been used for biodosimetry
(Vaurijoux et al. 2009; Vaurijoux et al. 2015). However, it is
still necessary to manually pre-process and supervise DC analyses
performed with this software. After cells with abnormal chromosome
counts and according to Metasystems, "metaphases where the two
chromatids are sticked or with twisted chromosomes, and metaphases
where centromeric constrictions are not visible" are removed, the
remaining images are analyzed with the software (Gruel et al.
2013). The operator then excludes images with "twisted chromosomes,
two aligned chromosomes, and other figures detected as dicentrics
by the software." False positive (FP) DCs will alter the estimated
dose if these steps are not performed (Gruel et al. 2013).
DCScore.TM. software is therefore considered semi-automated because
it requires manual pre- and post-processing review of DCs (Schunck
et al. 2004; Vaurijoux et al. 2015), especially at low radiation
doses.
The high rate of FPs in raw data is not surprising considering the
known variation among chromosome morphologies. The detection of
DCs, which are much less frequent than monocentric chromosomes
(MCs), is also impacted by differences in sample processing
procedures among laboratories (Wilkins et al. 2008).
One challenging issue with automated analysis is the selection of
images of adequate quality for accurate identification of the
chromosome damage. Selection of images for cytogenetic biodosimetry
has traditionally requires a subjective, manual review of images to
determine those of sufficient quality to score DCs unequivocally.
The decision to manually select or exclude microscope images for
DCA has not been based on analyses of quantitative analysis of
structural properties in each chromosome, metaphase cell, or the
level of contamination with non-chromosomal objects; without
attention to this image properties, automated cell image capture
approaches make this approach impractical due to the growing size
of datasets of many samples consisting of thousands of images.
Image quality assessment often estimates new data in relation to
reference images (Zhou et al. 2004), complex mathematical models
(Nill and Bouzas 1992), or distortions from a training set
recognized by machine learning (Narwaria and Lin 2010). Generic
methods of assessing image quality are not appropriate in our
situation. Features tailored for ranking chromosome images cannot
be generalized to entropy measures based on applying frequency
filter to intensity distributions. To be useful, quality assurance
for evaluation of specific microscopic biological objects in an
image may require expert-derived rules to categorize preferred
images. When performed manually, the speed of this process can vary
significantly between samples, and the accuracy may reflect
subjective evaluation between experts or may not be reproducible
with the same expert. Improved methods were sought that uniformly,
reproducibly, and automatically evaluate the suitability of
metaphase images, since these improves the accuracy of dose
estimation. Furthermore, implementation of this automated process
provides feedback control of the microscope system that enables
samples to complete processing once the system indicates that a
sufficient number of high quality metaphase images have been
ascertained.
SUMMARY
The methods, systems, and platforms of the present disclosure
provide means for automatically controlling a microscope system for
determining exposure levels to ionizing radiation using the DCA.
This smart microscopy system processes images of metaphases
chromosomes and classifies DCs, if any, in each image. Then, the
system selects those images with the most accurate identifications
of DCs, and determines radiation exposure levels in biological
samples by comparison of system-generated radiation dose
calibration curves derived from the DCA. When a sufficient number
of high quality images have been ascertained from a sample for dose
estimation, the digitally-controlled microscope system terminates
acquisition of additional images by the microscope system, thus
expediting both data collection and interpretation. This is an
improvement over conventional digitally controlled microscopy
systems for cytogenetic biodosimetry analysis, since the instant
invention reduces the amount of time required to obtain sample data
while at the same time, ensures that cytogenetic biodosimetry
images meet the quality requirements for the assay.
Disclosed herein is an imaging platform comprising: (a) a
microscope system capable of recognizing and digitally imaging
metaphase chromosomes, and (b) a processor configured to perform
image analysis, wherein the image analysis comprises: the Automated
Dicentric Chromosome Identifier (ADCI) software to automate DC
scoring and radiation dose estimation. The algorithms underlying
ADCI have been described and experimentally validated (Li et al.
2016; Arachchige et al. 2010; Arachchige et al. 2012; Arachchige et
al. 2013; Subasinghe et al. 2016). Briefly, foreground objects are
extracted from the metaphase cell image by thresholding intensities
above background levels using a gradient vector flow method.
Preprocessing filters remove most (but not all) non-chromosomal
objects (e.g. debris, nuclei, overlapping chromosomes). Each
remaining object is regarded as a single, intact, post-replication
"chromosome" object. These can include objects that DCScore rejects
as possible DCs, specifically those chromosomes with separated
sister chromatids, where the individual chromatids are in close
proximity to one another and are tethered to the same centromere
and are therefore recognized as synapsed. In ADCI, each chromosome
is processed to determine a contour (chromosome boundary) and its
centerline (chromosome long axis) by discrete curve evolution. The
Intensity-Integrated Laplacian method (Arachchige et al. 2013;
Subasinghe et al. 2016; U.S. Pat. No. 8,605,981 constructs a width
profile from consecutive vector field tracelines running
approximately orthogonal to the centerline, and potential
centromere locations ("centromere candidates") are identified from
constrictions in the said width profile (see FIG. 1). Machine
learning (ML) modules use image segmentation features derived from
each chromosome to classify centromeres and dicentric chromosomes
(Li et al. 2016; Rogan et al. 2016). The first Support Vector
Machine (SVM) ranks potential centromere candidates in each
chromosome according to their corresponding hyperplane distances;
then another SVM scores the chromosome as either monocentric (MC)
or dicentric (DC) using features derived from the top two
candidates.
Samples exposed to known radiation doses (in Gy) are processed by
ADCI to construct a dose-response calibration curve. The average
frequency of DC's per cell in dose calibrated samples, the
radiation response, is fit to a linear-quadratic function. The
response for test samples exposed to unknown radiation levels can
then be analyzed with this equation to estimate their corresponding
doses.
We noticed that metaphase cell images of inconsistent quality can
affect accuracy of dose estimation by ADCI. Previous studies
evaluated the efficacy of ADCI at chromosome classification and
dose estimation.sup.10,11. While the sensitivity (recall) for DCs
was acceptable (.about.70%) and relatively constant at all
radiation exposure levels, precision showed a strong dependence on
dose. Chromosome misclassification, in particular false positive
dicentrics (FPs) were more prevalent at low (.ltoreq.1 Gy) compared
to high (3-4 Gy) doses; at 1 Gy, FPs could outnumber true positive
dicentrics (TPs) by a factor of 4 to 5. Consequently,
ADCI-processed samples exhibited a reduced range of accurate
responses to radiation compared to manually scored samples.
Although use of the same algorithm to derive the calibration curve
compensates for some of these differences, reliability of dose
estimation ultimately hinges on DC classification accuracy. As DCs
are greatly outnumbered by MCs (background frequency in normal,
unexposed individuals is one DC per 1000 cells), this study focuses
on improving the distinction between TP and FP DCs without
compromising recall.
FPs reflect inadequacies in misinterpreting certain chromosome
morphologies or non-chromosomal objects. Selective targeting and
removal of these instances would reduce FPs without limiting TP
identification, improving overall classification accuracy. We
investigated FP morphologies to identify problematic cases and
devised a set of post-processing object segmentation filters to
eliminate them. Then, to ensure consistent performance,
segmentation filters were developed to remove poor quality cell
images. These images are usually characterized by either a lack of
or incomplete complement of metaphase chromosomes, misclassified
interphase or micro-nuclei as metaphases, or incorrectly segmented
sister chromatids as individual chromosomes. Each proposed filter
was tested individually, and the best performing filters were
integrated, and tested on actual cytogenetic dosimetry data exposed
to various radiation doses. The effects of these filters on
classification performance was evaluated on image sets from two
independent biodosimetry laboratories, and their impact on dose
estimation was assessed on cells obtained from an international
biodosimetry exercise.
We present this approach which selects images based on a
combination of optimal global image properties for scoring
metaphase cells, and customized object segmentation, identification
and elimination of false positive DCs. Automated image selection
with these segmentation filters reduces the number of images that
are required to capture metaphase cells, thus decreasing the number
of images and time required to process each sample. The dynamic
reduction in the number of images that are acquired by the system
is a particular advantage of the instant invention over
commercially available systems which require that a fixed number of
images be captured by the system, regardless of quality. If fewer
images can be used to estimate radiation exposure, this will reduce
the amount of time required to obtain data for each sample. In a
moderate to large scale mass casualty, this time differential will
enable more samples to be processed by the system described here
than other commercial systems that do not dynamically assess image
quality in real time. The computer that performs this quality
assessment could be the same computer that drives the microscope
system to acquire metaphase cell images, but it is more likely that
a different computer will perform these calculations by accessing
newly captured images on a storage unit that is shared by both
machines on the same network, as higher performance hardware can
expedite the decisions regarding the suitability and quality of
images, thus minimizing the acquisition of unnecessary images by
the microscope system. Thus, the system may consist of multiple
computers performing different tasks, ie. the first computer to
direct and manage the motorized stage, objective turret and digital
camera attached to the microscope to collect images of metaphase
cells and a second computer performing the quality assessment of
the images acquired by the first computer. The second computer
communicates with the first computer when a sufficient number of
quality images have been obtained to terminate collection of data
on a sample and instruct the first computer to proceed to the next
microscope slide or sample. For example, ADCI has been found to
reject approximately 30% of all high magnification metaphase images
produced natively by Metafer (Metasystems) software in which the
default classifier is used to detect metaphase cells. Since 90% of
the time used to capture images per sample are detected during this
step of metaphase finding, the time required for instant invention
to capture all the requisite images in a sample will be 27% less
than if the system had not been used. A microscope system
integrates image selection procedures by electronically sending
instructions to an automated digitally controlled microscope to
discontinue collecting images after it determines that sufficient
number of high quality metaphase cell images have been acquired to
detect DCs accurately. The novel aspect of this software is that
the microscope system will accurately determine radiation dose from
these images, while terminating data collection sooner than it
would have been normally programmed to end this process. These
improvements in the ADCI-controlled microscope system ensure
timely, reproducible, and accurate quantitative assessment of acute
radiation exposure.
In some embodiments of the disclosed digitally controlled
microscope system with a motorized stage to move between cells,
motorized turret to change optical objectives, executable software
performs segmentation of chromosome objects in microscope images,
determines image quality and counts the number images of sufficient
quality for cytogenetic analysis, then determines whether
chromosome objects are either monocentric, dicentric chromosomes or
neither of these objects, and counts the number of DCs in each
image, then determines whether the number of images and dicentric
chromosomes fulfill criteria published by the IAEA (2011) for
performing the DC assay. If these criteria met, the system
determines the frequency of DCs over all images from the same
individual, estimates the exposure to radiation based on this
frequency.
In some embodiments of the disclosed microscope system with a
digital camera, a digital processor configured to perform
executable instructions that analyze images of cells produced by
said camera, and instruct the microscope system to continue or
terminate the collection of images, which the microscope system
then performs. Upon termination the data collection and processing
of this sample, the aforementioned process is repeated for the next
sample, if any.
In some embodiments disclosed digitally controlled microscope
system, executable software processes chromosome objects present in
the images generated by the microscope system, and once all of the
chromosome objects from the same sample are processed, the system
calculates the radiation exposure (in Gy) of a sample based on the
DC frequency in a set of metaphase images from a sample, including
either whole body or partial body exposures.
In some embodiments, executable software also calculates a
confidence interval for the dose of radiation exposure.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1. Chromosome images processed by ADCI, annotated with key
segmentation features. (A) Monocentric and (B) Dicentric
chromosome. Chromosome contour is overlaid in green, long-axis
centreline in red. Yellow and cyan markers on the centerline
indicate the top-ranked and 2.sup.nd-ranked centromere candidate,
respectively (other candidates not shown), with their corresponding
width tracelines (roughly orthogonal to centerline) displayed in
the same colour. Arc lengths of width tracelines running down the
centerline (not all shown) are used to construct a chromosomal
width profile. Note that the top-ranked candidate correctly labels
the true centromere location, while the 2.sup.nd-ranked candidate
labels a minor non-centromeric constriction. By comparing features
extracted from both candidates (including width and pixel intensity
information), the software correctly assessed that only one of the
candidates is an actual centromere, so the chromosome was
classified as monocentric. In dicentric chromosomes, both
candidates would label actual centromeres.
FIG. 2. Examples of FPs in each morphological subclass. The
subclasses are defined in the Methods 2. Chromosome contours are
displayed in green, centerlines in red, top-ranked and
2.sup.nd-ranked centromere candidates in yellow and cyan,
respectively, and other centromere candidates in blue. (A) SCS: An
MC with SCS showing the characteristic localization of centerline
along chromatid. (B) Chromosome fragment: Artifactual fragmentation
of a chromosome caused by overaggressive image segmentation. (C)
Chromosome overlap: Two touching MCs treated as a single DC
(under-segmentation). (D) Noisy contour: The jagged contour due to
poor image contrast is prone to introducing artifactual width
constrictions. (E) Cellular debris: Incorrectly processed as a
chromosome. (F) ML deficiency: An MC with no notable errors in
contour or centerline.
FIG. 3. A visualization of DC filter scores for a particular FP. DC
Filters are defined in Methods 3.1. (A) A processed FP (acrocentric
chromosome with SCS), with contour in green, centerline in red,
top-ranked centromere candidate and its width traceline in yellow,
2.sup.nd-ranked centromere candidate and its width traceline in
cyan. (B) Filter I: Thresholded binary image of the chromosome is
used to calculate pixel area (in white). (C) Filters II-V: Width
profile along centerline is shown in red (horizontal axis plots
centerline location, vertical axis plots width), with mean width in
green (filter II), median width in blue (filter III), max width in
magenta (filter IV), and width of top centromere candidate in
yellow (filter V). (D) Filter VI: Contour in blue and its minimum
bounding rectangle in magenta and green. (E) Filter VII:
Partitioning of contour at centerline endpoints (intersection of
red line with contour) into two segments, green and blue. (F)
Filter VIII: Traceline endpoints of top 2 centromere candidates
(intersection of yellow and cyan lines with contour) are used to
partition contour into 4 segments (1 blue, 1 green, 2 magenta);
relative arc lengths of blue and green segments are taken into
consideration.
FIG. 4. Cell image viewer in ADCI demonstrating example of a
corrected FP DC. Graphical User Interface for viewing cell images
within a sample processed by ADCI.sup.11. Valid segmented objects
(generally chromosomes, but occasionally nuclei or debris) are
shown with coloured contours. Red contours indicate predicted DCs,
yellow contours indicate chromosomes that were initially classified
as DC but removed by the FP filters (new), green contours indicate
predicted MCs, and blue contours indicate objects that could not be
further processed after segmentation. Beneath the image, new
controls were added to allow manual inclusion/exclusion of images
within a sample from dose analysis.
FIG. 5. Calibration curves for HC and CNL samples. The
dose-response calibration curves for (A) HC and (B) CNL metaphase
cell image sample data. Response (mean DC frequency) on vertical
axis, corresponding radiation dose (Gy) on horizontal axis. Green
curves are based on unfiltered images, cyan curves were derived by
recomputing DC frequencies after applying FP filters (filters
I+IV+V+VI+VIII) to these datasets. HC curves are constructed by
fitting a linear-quadratic curve through all HC calibration
samples, CNL curves are similarly constructed from CNL calibration
samples (refer to Table 2). The CNL curves consistently show a more
pronounced quadratic component than the HC curves, which exhibit a
nearly linear response. After applying FP filters (cyan), the
curves show a diminished dose-response (green), due to elimination
of some detected FP DCs.
FIG. 6. Original vs. manually curated calibration curves for HC
samples. The dose-response calibration curves for HC sample data,
with and without FP filters applied, before and after curation.
Response (mean DC frequency) on vertical axis, corresponding
radiation dose (Gy) on horizontal axis. Green curve is not curated
with all images included, cyan curve is not curated with FP filters
applied, red curve is curated but unfiltered, and blue curve is
curated and FP filters have been applied. Uncurated curves were
generated from 0, 0.5, 1, 2, 3 and 4 Gy calibration image data (see
Table 2). Curated curves were generated from the same data (however
0.5 Gy was not included) after lower quality images were manually
removed (see Methods 6). After manual curation, the curves show a
stronger quadratic component, similar to the CNL curves (see FIG.
5).
FIG. 7. Relation between DC frequency (y-axis) and number of
included top images (x-axis) when images are ranked by different
scoring methods, in sample HC3 Gy. Blue, orange and green curves
correspond to unordered images (alphabetic order of image names),
images sorted by group bin method and images sorted by combined
z-score method, respectively. Figure was generated using Plotly
software.
FIG. 8. The structure of UML diagram. Panels show: (A) Two new
wizard page classes (search space, evaluation) are required.
Selection model generation dialog is a new class. The dialogs
display three panels to i) enter samples and specify parameters and
model evaluation methods, ii) show progress bar, and iii) displays
results after selecting best models. Besides the user interface
members, it implements the process to find optimal image selection
models. The Wizard reuses class `Sample`, `ImageSelectionMode` and
`LinearModel`. (C) Presentation of selection models generation
result is part of selection model generation dialog, or might have
its own class. Reuse class `ImageSelectionDialog`.
FIG. 9. Summary of model generation configuration and evaluation
method and the samples to be evaluated, and display of selected
model parameters upon completion of the search.
DETAILED DESCRIPTION OF THE INVENTION
Cytogenetic data were obtained by biodosimetry laboratories at
Health Canada (HC) and Canadian Nuclear Laboratories (CNL)
according to IAEA guidelines. Blood samples were irradiated by an
XRAD-320 (Precision X-ray, North Branford, Conn.) at Health Canada
and processed at both laboratories. Peripheral blood lymphocyte
samples were cultured, fixed, and stained at each facility
according to established protocols (Wilkins et al. 2008; IAEA
2011). Metaphase images from Giemsa-stained slides were captured
independently by each lab using an automated microscopy system
(Metasystems). One set of metaphase images from CNL and two sets
from HC (Table 1) were used for development and initial testing of
the proposed algorithms. After image processing by ADCI, called DCs
were manually reviewed and the consensus scores of TPs or FPs by 3
trained individuals were determined. Calibration curves were
prepared based on 6 samples of known radiation dose (Table 2). An
additional 6 samples.sup.11 were initially blinded to the actual
radiation exposures as test samples (Table 3). Test samples were
exposed to a range of radiation doses bounded by the doses of
samples used to construct the calibration curve. The sample naming
convention is the lab name followed by the sample identifier, e.g.
HC1 Gy signifies the 1 Gy calibration sample prepared at HC,
whereas CNL-INTC03S04 represents the INTC03S04 international
exercise test sample (exposed to 1.8 Gy) prepared at CNL.
Data consisted either of all "metaphase" images captured by the
microscopy system, or a manually curated set of 500 high quality
images. Selection of raw metaphase images for inclusion in samples
was done automatically at HC using the default image classifier of
the Metafer slide scanning system, while CNL selected images
manually according to IAEA guidelines. Experts from CNL selected
for images deemed analyzable by humans with respect to chromosome
count, spatial distribution and morphology.
1) ADCI Settings & Metaphase Image Data
ADCI software (V1.0) was used for DC detection and dose prediction,
with the MC-DC SVM tuning parameter, .sigma., set to 1.5. ADCI
libraries were initially written in MATLAB (R2014a) to develop and
test the proposed DC FP filters, and were subsequently rewritten in
C++ and integrated into ADCI. For development and validation of
segmentation filters, independent datasets used three sets of
roughly 200 images each (2 low dose, 1 high dose) were prepared
from larger image sets that were originally used for validation of
previous versions of ADCI (see Table 1; HC-mixed image set).
2) Morphological Characterization of FPs
FPs and TPs were compared according to their respective
segmentation features, including contour, width profile, centerline
placement, centromere candidate placement, and total pixel area
(Table 1). FPs were grouped by common distinguishing traits and
assigned to one or more of the following morphological classes:
I. Sister Chromatid Separation: Sister chromatid separation (SCS)
of a chromosome refers to the loss of sister chromatid cohesion at
the telomeres, and often along the sister chromatids, excluding the
centromeres. Due to inherent limitations of a centerline derived
from contour skeletonization in chromosomes, SCS often resulted in
partial or complete localization of the centerline along a single
chromatid, rather than along the long axis of the full-width
chromosome (Arachchige et al. 2012; Arachchige et al. 2013;
Subasinghe et al. 2016). Complete centerline localization to
chromatids of the q arm was common among acrocentric chromosomes
(see FIG. 2A). This resulted in a width profile in which the
displaced centerline did not accurately represent the width of the
chromosome, and compromised centromere determination.
II. Chromosome Fragmentation: Sister chromatid pairs were
completely dissociated in metaphase images, resulting in incorrect
labeling of each chromatid as separate chromosomes. Occasionally,
segmentation fragmented images of intact non-uniform chromosomes
into multiple, chromosomal artifacts.sup.6 (see FIG. 2B).
Artifactual fragmentation into incomplete chromosome fragments led
to unpredictable results, increasing FPs and FNs.
III. Chromosome Overlap: Poor spatial separation of chromosomes
produced clusters of overlapping/touching chromosome clusters which
were inseparable. Occasionally, the cluster is segmented as a
single contiguous object (see FIG. 2C). Like chromosome fragments,
analysis of these overlapping chromosome clusters produces
erroneous results. FP DCs were produced from clusters comprising
two underlying monocentric chromosomes, each contributing a
centromere to the combined object.
IV. Noisy Contour: Poor image contrast at the chromosomal boundary
produced "noisy," jagged chromosome contours contributing multiple
small constrictions to the width profile (see FIG. 2D). These
artifactual constrictions were incorrectly identified as multiple
centromeres if their magnitudes were similar to the true
centromere, leading to FP assignment.
V. Cellular debris: Non-chromosomal objects such as nuclei and
cellular debris were generally removed by pre-processing based on
thresholding relative size and pixel intensity. However, aggregated
cellular debris were occasionally labelled as a chromosome and
naively analyzed by the software (see FIG. 2E).
VI. Machine Learning Error: A "catch-all" subclass for MCs with no
identifiable morphological traits and reasonable contours and
centerlines (see FIG. 2F). These cases reflect deficiencies in the
feature set or training data of the machine learning (ML)
classifiers, rather than image segmentation errors.
3) Filtering Out False Positive Objects
Quantitative filters were created and tested to delineate FP DCs.
Each formula targets one or more of the morphological classes
described above, and generates a unitless filter score for each
object, independent of the biodosimetry reference laboratory
source. For any metaphase image, {c.sub.1, . . . , c.sub.N} denotes
the set of N chromosomes within the image and c* denotes the
predicted DC of interest. Each filter classifies c* as either a TP
or FP by comparing its filter score against a heuristically-defined
threshold that is independent of laboratory provenance. Thresholds
were established empirically to maximize elimination of FPs without
altering recognition of TPs. FPs generally produce lower filter
scores than TPs (i.e. lower area, lower width, less oblong
footprint, more asymmetrical), so FPs were selectively targeted by
eliminating candidate DCs with scores below a threshold. Due to the
low frequency of DCs in any given sample, minimizing the loss of
TPs is paramount to minimize the likelihood of TP removal. For each
filter, corresponding filter scores were calculated for all DCs in
the HC-mixed image set (Table 1), and a heuristic threshold (to 2
significant digits; see below) was set to the minimum value
observed in TPs. Thresholds for filters VI to VIII were calculated
by repeating the same procedure on a chromosome set of 244 TPs from
the MC-DC SVM training set, and the final thresholds were set to
the lower of each pair of values.
I. Area Filter: A(c) denotes the pixel area occupied by chromosome
c (see FIG. 3B). c* was classified as FP if
A(c*)/median({A(c.sub.1), . . . , A(c.sub.N)})<0.74 or as TP
otherwise. This filter targets small acrocentric chromosomes
(commonly displaying SCS) and chromosome fragments.
II. Mean Width Filter: W.sub.mean(c) denotes the mean value of the
width profile of chromosome c (see FIG. 3C). c* was classified as
FP if W.sub.mean(c*)/median({W.sub.mean(c.sub.1), . . . ,
W.sub.mean(c.sub.N)})<0.80 or as TP otherwise. This filter
targets SCS and chromosome fragments.
III. Median Width Filter: W.sub.med(C) denote the median value of
the width profile of chromosome c (see FIG. 3C). c* was classified
as FP if W.sub.med(c*)/median({W.sub.med(c.sub.1), . . . ,
W.sub.med(c.sub.N)})<0.77 or as TP otherwise. This filter
targets SCS and chromosome fragments.
IV. Max Width Filter: W.sub.max(c) denotes the maximum value of the
width profile of chromosome c (see FIG. 3C). c* was classified as
FP if W.sub.max(c*)/median({W.sub.max(c.sub.1), . . . ,
W.sub.max(c.sub.N)})<0.83 or as TP otherwise. This filter
targets SCS and chromosome fragments.
V. Centromere Width Filter: W.sub.cent(c) denotes the width of
chromosome c at the top-ranked centromere candidate (see FIG. 3C).
c* was classified as FP if
W.sub.cent(c*)/median({W.sub.cent(c.sub.1), . . . ,
W.sub.cent(c.sub.N)})<0.72 or as TP otherwise. This filter
targets SCS and chromosome fragments.
VI. Oblongness Filter: S(c) denotes the pair of side lengths of the
minimum bounding rectangle enclosing the contour of chromosome c
(see FIG. 3D). c* was classified as FP if
1-min(S(c*))/max(S(c*))<0.28 or as TP otherwise. This filter
targets acrocentric chromosomes with SCS and some cases of
overlapping chromosomes.
VII. Contour Symmetry Filter: Let L(c) denote the pair of arc
lengths of contour halves produced by partitioning the contour of
chromosome c at its centerline endpoints (see FIG. 3E). Classify c*
as FP if min(L(c*))/max(L(c*))<0.51 or as TP otherwise. This
filter targets SCS.
VIII. Intercandidate Contour Symmetry Filter: L.sub.C(c) denotes
the pair of arc lengths of the contour regions of chromosome c that
run between the traceline endpoints of its top 2 centromere
candidates (see FIG. 3F). c* was classified as FP if
min(L.sub.C(c*))/max(L.sub.C(c*))<0.42 or as TP otherwise. This
filter targets SCS and some cases of overlapping chromosomes.
Incorporation into Existing Algorithms:
After chromosome processing and MC-DC SVM classification.sup.11 but
prior to dose determination, all DC chromosomes inferred by ADCI
were analyzed with the proposed DC filters. DC filter scores
exceeding TP thresholds were included in the dose determination,
whereas DCs classified as FPs by any filters (inclusive "or") were
eliminated. DCs that were filtered out are outlined in yellow in
the ADCI cell image viewer.sup.11 (FIG. 4).
Determination of Optimal Filter Subset:
The proposed filters were not completely independent of each
another, as some measures were related to the same chromosome
segmentation features (i.e. width for filters II-V, contour
symmetry for VII-VIII) and/or targeted the same morphological
subclass (notably SCS). Thus, the "optimal" filter subset (termed
"FP filters") was defined as the subset of filters which maximized
FP removal ability while minimizing redundant FPs. Performance for
a given set of filters was the total percentage of FPs removed by
any of its filters (inclusive "or") in the HC-mixed image set (see
Table 1). Using a forward selection approach, individual filters
were added iteratively to identify those which produced the largest
improvement in performance.
Evaluation of FP Specificity on HC Test Samples:
All objects removed by the FP filters in each image in HC samples
INTC03S01, INTC03S08 and INTC03S10 (Table 3) were manually reviewed
(FIG. 4). Filtered TPs and filtered objects with ambiguous
classifications (TP or FP) were reviewed with another expert before
final classification. For each sample, the number of filtered FPs
was determined by subtracting number of filtered TPs from the total
filtered count, and FP specificity was defined as the ratio of
count of FPs to that of all filtered objects.
4) Dose Estimation Analysis
In ADCI, a pre-computed dose-response calibration curve is also
used to estimate radiation absorbed in samples with unknown
exposures.sup.11. For a given sample, ADCI calculates the mean
response from total number of detected DCs divided by the number of
cell-containing images. Calibration curves can be generated from a
set of calibration samples either by processing and calculating a
response for each sample, or allowing the user to input the
corresponding response, and fitting the dose-response paired data
to a linear-quadratic curve by regression. Because sample
preparation protocols vary between laboratories, dose estimation of
test samples were performed with calibration curves generated by
the same source.sup.11.
Distinct calibration curves were generated for each laboratory,
either enabling or disabling FP filters, for the 0, 0.5, 1, 2, 3
and 4 Gy calibration samples (see Table 2). Radiation doses of
images obtained by HC for test samples (Table 3) were estimated
using the HC calibration curve derived by ADCI after applying the
same FP filters. A similar analysis was carried out for the 5 CNL
test samples using the CNL calibration curve data.
5) Effect of Filtering on Manually Image Selected HC Data
To investigate the impact of manual image selection on dose
accuracy, we compared HC calibration curves derived from manually
curated samples with the FP filters either enabled or disabled
(Table 2). Manual curation of the HC samples was similar to manual
image selection performed by CNL. Images were selected requiring:
I) Complete complement of approximately 46 chromosomes, >40
segmented objects, <5 segmented objects from different nuclei if
multiple nuclei present; II) Exclusion of "harlequin" chromosomes.
Cells with unevenly stained sister chromatids cultured in the
presence of bromodeoxyuridine (BrdU), which is indicative of
2.sup.nd division metaphases, were excluded (Subasinghe et al.
2016); III) Well-spread, sharply-contrasted chromosomes with
minimal sister chromatid dissociation. Only images with <5
incorrectly-segmented chromosomes were included, where incorrect
segmentation was defined as chromosome overlaps (indicating poor
spread), fragments (indicating sister chromatid dissociation) and
overly-noisy contours (indicating poor image contrast); IV)
Adequate chromatid condensation. Depending on the stage of
metaphase arrest, the degree of chromosome condensation can differ
(Rieder and Palazzo 1992; Sethakulvichai et al. 2012). Prometaphase
cells have longer chromosomes, are less rigid, exhibit greater
overlap and less well-defined centromere constrictions, all of
which pose a significant challenge for automated chromosome
classifiers (Sethakulvichae et al. 2012; Carothers and Piper 1994).
Metaphase images with longer, thinner chromosomes (roughly
corresponding to >500-band level) were also excluded. Guidelines
I-III and a minimum sample size of 500 cells were adopted from IAEA
recommendations, whereas guideline IV was added after preliminary
inspection of HC calibration samples. Manual curation was performed
within ADCI by retrospectively excluding images in processed
samples from dose analysis (FIG. 4). For each sample, consecutive
images meeting all criteria were evaluated until 500 images were
accrued. DC classifications were hidden during image selection to
minimize bias. After generation of the curated HC calibration
curves, the radiation doses of the three HC test samples (Table 3)
were re-estimated on the new curves, with and without the FP
filters enabled.
6) Automating Removal of Suboptimal Images by Morphology
Filtering
Reference biodosimetry laboratories screen for interpretable
metaphase cell images prior to DC analysis. Manual selection of
images assures consistency and reliability of metaphase data, which
increases analytic accuracy. As automated DC analysis can also be
affected by variable cell image quality, excluding undesirable
images in a sample would be expected to reduce FPs, and expected to
more accurately estimate radiation exposures.
Image segmentation filters used empirically determined criteria to
eliminate metaphase cells with characteristics that increased FP
DCs. Image-level segmentation filters that threshold features I and
II (below) were used to detect cells in prometaphase (relatively
long and thin chromosome morphology), prominent sister chromosome
dissociation, and highly bent and twisted chromosomes; another
filter (III) detected overly-smooth contours characterized by
images containing intact nuclei and otherwise incomplete chromosome
sets. The total object count (IV) and segmented object count
filters (V) fulfill general criteria for nearly normal metaphase
images of approximately 46 chromosomes. These filters are used to
exclude images with extreme object counts. Filter VI selects images
based on effectiveness of chromosome recognition by ADCI.
Image level filters are calculated in terms of their z-scores of
all objects in an image. For any particular metaphase image I* in a
sample containing M images, {I.sub.1, . . . , I.sub.M}, where
{c.sub.1, . . . , c.sub.N} denote the set of N chromosomes within
image I*. Additionally, SD denotes the standard deviation function,
and T denotes the threshold SD common to all 3 filters that
identifies outlier images. This SD value was set heuristically to
1.5 after by varying T after applying these filters to the HC2 Gy
calibration sample (Table 2). Similarly, suggested thresholds in
filters IV-VI are also derived from experiences of testing multiple
samples. I. Length-width ratio filter (LW) defines the average
length-width ratio of chromosomes in an image. For a given
chromosome c in a given image I containing N chromosomes, L(c,I)
denotes the arc length of the centerline of c, and W.sub.mean(c,I)
denotes the mean value of the width profile of c. MW(I) is defined
as mean{L(c.sub.1,I)/W.sub.mean(c.sub.1,I), . . . ,
L(c.sub.N,I)/W.sub.mean(c.sub.N,I)}. 1* is removed if
MW(I*)>mean{MW(I.sub.1), . . . ,
MW(I.sub.M)}+T.times.SD{MW(I.sub.1), . . . , MW(I.sub.M)}. II.
Centromere candidate density filter (CD) counts occurrences of
centromere candidates in chromosomes. It eliminates images
containing chromosomes with a high density of centromere
candidates. For a given chromosome c in image I containing N
chromosomes, L(c,I) denotes the arc length of the centerline of c,
and N.sub.cent(c,I) denotes the number of centromere candidates
along c. CD(I) is defined as the
mean{N.sub.cent(c.sub.1,I)/L(c.sub.1,I), . . . ,
N.sub.cent(c.sub.N,I)/L(c.sub.N,I)}. I* is removed if
CD(I*)>mean{CD(I.sub.1), . . . ,
CD(I.sub.M)}+T.times.SD{CD(I.sub.1), . . . , CD(I.sub.M)}. III.
Contour finite difference filter (FD) represents contour smoothness
of chromosomes in an image. It eliminates images with prominent
non-chromosomal objects with smooth contours, such as nuclei or
micronuclei. For a given chromosome c in a given image I containing
N chromosomes, WP.sub.D(c,I) denotes the set of first differences
of the normalized width profile of c (range normalized to interval
[0,1]). WD(I) is defined as the
mean{mean{abs{WP.sub.D(c.sub.1,I)}}, . . . ,
mean{abs{WP.sub.D(c.sub.N,I)}}}. I* is removed if
WD(I*)<mean{WD(I.sub.1), . . . ,
WD(I.sub.M)}-T.times.SD{WD(I.sub.1), . . . , WD(I.sub.M)}. IV.
Total object count (ObjCount) filter defines the number of all
objects detected in an image. Values lying outside of a threshold
range are rejected to eliminate images with multiple metaphases or
excessive cellular debris. Based on empirical analyses, the
suggested object count range falls within the interval [40, 60]. V.
Segmented object count (SegObjCount) filter defines the number of
objects processed by GVF algorithm in an image. It is applied in
the same way as filter IV. The suggested range for the object count
interval is [35, 50]. VI. Classified object ratio (ClassifiedRatio)
filter defines the ratio of objects recognized as chromosomes to
the total number of segmented objects. It prevents images in which
ADCI fails to process most chromosomes from being included. An
image is removed if the value is less than a threshold of either
0.6 or 0.7, which is determined by the desired level of stringency
for application of this filter.
Combining filters. Applying these filters sequentially to the same
image distinguished the metaphase images for dose estimation from
less optimal cells with increased FPs. This was done by combining
the Z-scores of the image filters in a linear expression of
features I-VI that provides an assessment of image quality. The
resultant total score represents the degree to which a particular
image deviates from the population of images in a sample:
Score=w(LW).times.z(LW)+w(CD).times.z(CD)-w(FD).times.z(FD)+w(ObjCount).t-
imes.|z(ObjCount)|+w(SegObjCount).times.|z(SegObjCount)|-w(ClassifiedRatio-
).times.z(ClassifiedRatio)
Each feature has a positive free parameter, weight, to adjust its
contribution to the total score. The term LW determines that longer
and thinner chromosomes in the image will increase the score, as do
bending and twisted chromosomes due to the term CD. Lower
chromosome concavity also drives the score higher because of FD
term. Object count and segmented object count describe chromosome
positioning, separated sister chromatid level, etc. Assuming the
majority of images in a sample are good images, these terms will
result in higher scores for images exhibiting either incomplete,
multiple cells or severe sister chromatid separation. The last
terms produce high scores for images that the algorithm does not
process accurately. Images with smaller combined score are of
higher quality. The weights used are identified by evaluating many
possible weights and selecting those that minimize the error in
curve calibration. The weights obtained are optimal for calibration
samples, which will perform well on test samples, subject to the
condition that the calibration and test samples have comparable
chromosome morphologies. The score, however, cannot be used for
inter-sample image quality comparisons, as z-scores are normalized
within a sample.
Another, more general method was also developed to assess metaphase
images separately from other images in the same sample. Image
morphology is the primary consideration in assessing metaphase
image quality. The most common problems in poor quality metaphase
cells are severe sister chromatid separation, excessive chromosome
overlap, fragments of chromosomes in image segmentation, and
multiple cells or incomplete cells in the same image. They result
in changes in either the number of objects or areas of objects. For
instance, chromatid separation and chromosome fragments cause more
objects to be present in an image while areas of some objects are
smaller than normal. Chromosome-overlaps reduce the number of
objects, but their areas exceed those of discrete chromosomes.
To derive this novel quality measure, we exploited the general
property that the different chromosome lengths are approximately
proportionate to the known base-pair counts of each complete human
chromosome. By comparing the distribution of observed chromosome
object lengths with the gold standard derived from the lengths
obtained from the human genome sequence, we can assess the overall
quality chromosome segmentation of each cell. This assumption sets
aside chromosome abnormalities which result from radiation
exposure, which will be distributed randomly among cells analyzed,
because the cells are synchronized and harvested after a single
division. The actual chromosome lengths are difficult to measure
accurately in images, so instead, individual chromosomes are
approximated according to their corresponding chromosome areas (in
pixels). Therefore, the area of an object in a metaphase image is
used as a surrogate for which chromosome it represents. Once noisy
non-chromosomal objects, nuclei and large overlapped chromosome
clusters are removed, areas of the remaining objects are then
calculated based on their fractions to the total area of all
chromosomes, as overlapping chromosomes and chromatid separation do
not significantly affect the total area of objects in each
metaphase image. We bin the chromosomes in metaphase cell into
three categories corresponding to the known cytogenetic
classification system.sup.16: group A and B (AB), group C (C) and
groups D, E, F, and G (DG). A chromosome in category AB contains
more than 2.9% (determined by the shortest B group chromosome) of
total base-pairs in the complete chromosome set. A chromosome in
category C has less than 2.9% (determined by the longest C group
chromosome) but more than 2% (determined by the shortest C group
chromosome) of total base-pairs in the set. Any chromosome in
category DC contains fewer than 2% (determined by the longest D
group chromosome) of the total base-pairs. These thresholds 2.9%
and 2% are acceptable for the X and Y chromosomes, respectively. We
apply these thresholds to object areas to count the number of
chromosomes in each category in a metaphase image. An ideal
metaphase image will have 10 AB chromosomes, 16 C chromosomes and
20 DG chromosomes if the individual is female, or 10 AB
chromosomes, 15 C chromosomes and 21 DG chromosomes if the
individual is male. Images with chromosome overlap will tend to
have increased AB chromosome counts, while images with sister
chromatid separation will likely have elevated DG chromosome
counts. The morphological quality of a metaphase image can be
measured by comparing its chromosome categorizing result to the
female/male standard. In practice, we treat the categorizing result
of an image as a 3-element vector and calculate the Euclidean
distance to the standard. A larger distance corresponds to a less
satisfactory image, and we find that this measurement is universal
for metaphase images from different samples.
When images in a sample are sorted, by either combined z-score or
by chromosome group bin area measurement, a certain number of top
ranked images can then be selected for dicentric chromosome
analysis. Complex image selection models can be created by
filtering images first with filters and then selecting a certain
number of top scoring images.
7) Sample Quality Confidence Measurement
Metaphase image artifacts such as sister chromatid separation and
chromosome fragmentation interfere with the ability to correctly
identify dicentric chromosomes and compromises the reliability of
dose estimates. This dependence of dose estimation accuracy on
sample image quality motivates objective tests to evaluate and flag
data from lower quality samples and exclude such images from
analysis. Samples exposed to low LET whole-body irradiation,
typically seen in radiation incidents, exhibit DCs frequencies that
follow a standard Poisson distribution (IAEA 2011) of DCs per cell.
Deviation from the expected Poisson distribution can thus be
attributed to failure to accurately recognize and account for DCs
within the sample or by artifacts in the sample that are not
eliminated by the software. Following this principle, we devised a
sample quality evaluation method based on the conformity of the DC
count frequency distribution in each sample to a theoretical
Poisson distribution, as follows:
The number of DC occurrences in a cell is constructed as a
probability model in a sample. It is a discrete statistical model
as the number of events can only be integers. The appearance of any
DC is assumed to be independent of other DCs that may form. The
rate at which DCs occur is constant for a single sample at a given
radiation dose for full-body irradiation. The model of DCs per cell
detected by ADCI can therefore be approximated by a Poisson
distribution. The Poisson .lamda. parameter is obtained from the
average number of DCs per cell in a sample.
The observed DC distribution detected by ADCI is compared with the
Poisson distribution using the Pearson chi-squared goodness-of-fit
test. The test indicates the probability of observing the observed
data under the null hypothesis that they are Poisson distributed.
Samples without at least 1 cell image having >1 DC cannot be
analyzed, due to insufficient degrees of freedom. A smaller p-value
means the hypothesis is less likely and that the DC detection
results for that sample are less reliable. Very low p-values at or
below .alpha.=0.01 (99% confidence level) reject the null
hypothesis and indicate low quality samples.
Automated Determination of Optimal Image Selection Models
The foregoing describes methods of applying different image
segmentation parameters that select a subset of high quality images
collected by the microscopy system which provide improved accuracy
in the estimates of radiation dose for a specific sample. This
described process derived these parameters by empirical approaches.
Automatic generation of optimal image selections for given samples
in ADCI is preferable to users of the system because they may be
more accurate than empirical methods, tailored to the specific
samples and data quality available, and are less time consuming and
labor-intensive. The automation image selection is a self-contained
component in ADCI, including user interfaces and processing
functions. The communication between ADCI and image selection
automation are: (1) the main window in ADCIDE starts image
selection automation; (2) the work space in ADCI sends opened
processed samples to image selection automation; and (3) image
selection automation may use plot output and/or console output to
display results. Automatic generation of optimal image selection
model will be referred as `selection model generation` in
forthcoming comments.
Selection model generation is unconditionally delivered as a
component of the ADCI system, which means it is not an optional
add-on element to ADCI. Selection model generation is triggered by
an action in `Wizards` menu in main window (Currently `Curve
Calibration` and `Dose Estimation` are in the menu). Selection
model generation includes a wizard to collect data and parameters,
and a dialog to perform optimal image selection model generation.
The Selection model generation wizard contains all pages found in
the other wizards (eg. dose estimation, curve calibration) in ADCI,
that is, the introduction and conclusion pages. Completing all of
the wizard steps, prefills the selection model generation
dialog.
The Selection model generation wizard contains a page for users to
select loaded previously processed samples and to specify the
corresponding physical exposure values for samples of known
radiation dose. These samples are used to evaluate the accuracy of
each possible selection model over the range of parameter values.
The Selection model generation wizard provides a page for users to
select a SVM sigma value (a measure that balances specificity and
sensitivity for dicentric chromosome detection) of each of samples
used to derived the model. The sigma value used applies to both the
samples used to create the Selection model and those used to
evaluate it. The Selection model generation wizard also has a page
for users to specify parameters used to create the search space of
all of the parameter combinations used to specify different
possible Selection models. The search space of image selection
models consists of combined z-score models (A_[CDE]) and group bin
plus filtering models ([BCD]_[AB]). The number of selected images
can also be varied, which is particularly useful for samples that
contain images with poor morphology chromosomes or for models of
high quality samples requiring fewer images, which would reduce the
time that the system requires to processing a sample. The Selection
model generation wizard contains a page for users to specify
parameters for selection model evaluation. Three different methods
for the evaluation of image selection models have been validated,
including: 1) p-values of Poisson fit of dicentric chromosome
distribution in every sample (ideal samples do not differ from a
Poisson distribution, so p>0.05 is preferable); 2) calibration
curve fit residual (requires at least 3 samples); and 3) leave-one
cross-validation of dose estimation (requires at least 4 samples).
The main dialog of selection model generation contains a panel
shows a summary of collected data, parameters and info. A button on
the screen is selected by the user to start generation and
evaluation of the models. The main dialog of selection model
generation indicates a progress bar which is incremented during the
generation of all Selection models. Once all Selection models have
been evaluated, the main dialog of Selection model generation
presents a panel showing the results with the top 10 optimal
selection models found during the search. Users may then directly
save the optimal selection models as files so that they may be used
in the future, eg. by applying them to new samples.
Microscope System Control
Once the ADCI software has determined that a sufficient number of
high quality images have been identified and IEAA criteria have
been met for cytogenetic biodosimetry (either >100 dicentric
chromosomes or >500 cells), it can direct the microscopy system
to discontinue image capture from that sample, then proceed to a
new sample. This is a significant advantage over other microscopy
systems, since the time per sample is minimized by avoiding
collecting images that are unnecessary for the radiation dose
determining. In a radiation mass casualty or nuclear accident in
which multiple samples may need to be analyzed but the number of
microscopy systems to process those samples is limited, the time
savings from excluding unnecessary metaphase images can be more
effectively utilized by increasing the throughput of sample
processing.
ADCI has been integrated with an automated microscopy system for
metaphase cell capture, Metafer (Metasystems Inc) to discontinue
collecting further metaphase images from a sample. When sufficient
numbers of high quality images have been obtained for radiation
dose estimation on a slide from a sample, ADCI creates a file that
is checked frequently by Metafer for one of 5 possible commands
that can cancel steps, metaphase searches or resume metaphase
searching on a slide. Alternatively, it can broadcast a message
through Windows Message Broadcasting to the active Metafer window
cancel a search for further metaphase cells on a microscope Metafer
is configured to externally cancel a search for metaphase cells by
the microscope system by checking the file ExtCancel.TXT in the
path of the executable (C:\MetaSystems\Bin) within 1000 msec the
ExtCancelInterval in the MFGeneral section of the Metafer.INI (the
initialization/configuration) file on the microscopy system. The
text file created by ADCI contains on the first line the command
CancelSlide, which ends the metaphase search, and assuming that
each slide contains cells from a different sample, the metaphase
images on subsequent slide (corresponding to the next sample) will
begin to be captured. This command triggers Alternatively, using
Windows Message Broadcasting, the parameter "WMOnlmgExpo" in
MFGeneral of Metafer.INI is set to 1. The message is a comma
delimited string containing the export format and the number of
images that ADCI has determined should be exported. This number is
fewer than the number of images that have collected by the
automated metaphase finding Metafer (Metasystems) software.
Advantages of the Invention
Automated biodosimetric methods aimed at detecting DCs can produce
incorrect assignments because the algorithms cannot capture the
full range of morphological variability inherent in chromosome
images of metaphase cells. Accuracy of radiation exposure estimates
using automated biodosimetry can be improved by image segmentation
and filtering methods that remove suboptimal metaphase cell images
and eliminate false positive DCs. This study implements and tests a
set of morphology-based filters to eliminate FP DCs and unsuitable
metaphase images for automated biodosimetry. Compared to results
generated by the previous version of ADCI.sup.11, inclusion of
these filters reduced FP DC rates by .about.55% across a wide range
of radiation exposure levels. Additionally, we showed that these
filters were highly specific for FPs in test image sets as well as
actual patient samples (97.7-100%, n=6). Overall, the FP filters
substantially improve DC classification accuracy.
This is because proposed segmentation filters successfully target
SCS and chromosome fragments. In particular, the intercandidate
contour symmetry filter is a very promising SCS detector,
individually eliminating 84% of all SCS-induced FPs in our test
dataset. It was noted that acrocentric chromosomes were
disproportionally susceptible to SCS-induced errors compared to
other chromosome types (69% of SCS cases despite making up only 22%
of human chromosomes). Given the rarity of acrocentric TP DCs (due
to width profile inaccuracies at the extreme ends of
chromosomes.sup.7-9), filters targeting acrocentric or small
chromosomes, in general (such as filters I and VI), can also be
useful for reducing SCS-induced FPs.
Certain FP subclasses were commonly targeted by multiple filters.
Redundancy among the segmentation features resulted in only subset
of the filters being required to maximize elimination of FPs.
Notably, filters II-V eliminated FPs based different definitions of
chromosome width. The final combination of FP filters consisted of
only 5 of the 8 originally proposed; however, it should be noted
that a combination of only the intercandidate contour symmetry and
max width filters achieved nearly the same level of FP detection in
the test sample dataset, with the other filters having incremental
benefit.
Scale-invariance is an obligate property for any object-level
filter, since chromosome structures may vary between cells,
individuals, and laboratory preparations. Scale invariance is also
necessary to control for pixel-based chromosome measurements
affected by condensation differences over the course of metaphase
and differences in optical magnification. This principle was
achieved by either using filter scores normalized to the median
"raw" score of all objects within the same cell image (i.e. filters
I-V), or in which scores were derived from ratios of two
pixel-based measurements (i.e. filters VI-VIII).
There were differences in accuracy between the manually and
automatically-selected images for dose estimation. For the
previously manually curated CNL and HC samples, the FP object
filters respectively reduced the average dose estimation error from
0.4 Gy to <0.2 Gy (with a maximum error of 0.4 Gy). This placed
the accuracy our software comfortably within the .+-.0.5 Gy
requirement for triage purposes.sup.17. However, applying the FP
object filters alone to unselected HC metaphase data did not
improve accuracy (average error increased by 0.15 Gy). Thus, FP
object filters alone did correct for inaccurate dose response
estimates in all cases.
Variable cell image quality in some samples contributed to this
source of error. Some unselected HC samples contained images with
high levels of SCS, which upon processing produced large numbers
incorrectly classified chromosome fragments. Image level filters
I-V targeted these fragments, however they were not excluded based
on their threshold values, because they comprised the predominant
morphology within these particular cells. For similar reasons,
object-level filtering was not suitable for elimination for removal
of prometaphase images containing high resolution chromosomes
(>800 band level). These observations suggested the need for
image-level filters to select low quality images for removal in
addition to the object-level filters.
Image quality is critical to the accurate DC detection. Manual
inspection and quality control is common practice in cytogenetics
and biodosimetry laboratories, but it is labor-intensive.
Image-level filtering was automated to address this problem. These
methods apply statistical thresholds to morphological features of
chromosomes and non-chromosomal objects throughout a metaphase cell
image. Image scoring methods select a defined number of top-ranked,
processed images for dose estimation. The combined z-score method
is a weighted sum of standard deviations below or above the mean
score of objects in an image for each of the filter, and indicates
relative image quality. The chromosome group bin method is a more
general criterion that is calibrated to relative chromosome lengths
(and area) in base pairs. ADCI evaluates the morphological
deviation of chromosome area and ranks cell images relative to that
expected from the standard, normal karyotype. These FP filtering
and image scoring methods, which are referred to collectively as
image selection models, can be applied either individually or in
combinations within ADCI.
The significant improvement in accuracy of DC frequency is
attributable to both FP elimination and image selection. Dose
estimation errors with suitable image selection models in test
samples consisting of at least 250 images are considerably reduced.
The estimates are within the +/-0.5 Gy window of the corresponding
physical doses for the majority of samples tested. The current
image selection models in ADCI generally provide reliable image
quality control without manual intervention.
Automated image selection aims to simulate manual image curation.
Experiments demonstrated that the proposed methods successfully
improve dose estimates in test samples. At this point, automation
does not quite achieve the same overall accuracy, especially for
samples of variable quality. The respective differences in dose
estimates, especially at exposures >2 Gy, are not significant.
Automating image selection using this smart microscopy system,
nevertheless, offers unique advantages over manual image selection
in terms of analytic uniformity and speed.
EXAMPLES
Example 1. Application of Chromosome Morphology Filters to Remove
FPs
False positive DCs (n=97) from a low dose set metaphase images were
classified to uniquely identify, and ultimately eliminate these
objects. Chromosomal morphological subclasses (FIG. 3) included
those exhibiting excessive sister chromatid separation (I, n=51),
fragmentation (II, n=10), overlap (III, n=17), noisy contours (IV,
n=5), cellular debris (V, n=4), and inaccurate recognition by the
centromere candidate.sup.10 and MC/DC.sup.6 machine learning
algorithms (VI, n=11).
Segmentation filtering criteria were applied to these images.
Scale-invariant filters were tested to determine thresholds that
selectively removed subclasses I-III without eliminating any TPs.
Of the 51 SCS cases, 35 involved short, acrocentric chromosomes.
FPs were distinguished from TPs based on either their lower
relative pixel area or width (filters I-V), substantially
non-oblong footprint (filter VI), or substantial contour asymmetry
across the centerline (filters VII and VIII). For filters I-V,
normalization to median scores of other objects in the same image
performed similarly to normalization to other measures of central
tendency (e.g. z-score, mean, and mode after binning scores). FPs
were eliminated for each morphological subclass (Table 4), with
most of the segmentation filters acting on the targeted subclass,
however, the effects of each filter were not exclusive to those
subclasses (Methods 3).
To evaluate individual filter performance, the percentage of FPs
removed by each filter was calculated for the HC-mixed image set
(Table 5). A two-sample Kolmogorov-Smirnov test (K-S) was also
performed for each filter (.alpha.=0.05) on the same data, where
one sample consisted of the filter scores of all TPs (n=183) and
the other sample consisted of the scores of all FPs (n=158). All 8
filters rejected the null hypothesis (Table 5), suggesting that FPs
can be discriminated from TPs using empirically-thresholded filter
scoring. Application of the intercandidate contour symmetry filter
(Methods 3.VIII) achieved the largest overall reduction of FPs
(44.9%), and eliminated the most SCS-induced FPs (43 of 51). The
max width filter (Methods 3. IV) yielded the next largest reduction
in FPs (27.8%) and was the most efficient filter for detecting
fragmentation-induced FPs (8 of 10).
Additional FPs were eliminated by combining multiple segmentation
filters (see Methods 3). Since individual filters were separately
thresholded to avoid elimination of TPs (see Methods 3.2), the
inclusive disjunction (logical "or" operation) of multiple filters
had a negligible impact on TPs, while producing a stronger FP
discriminator. Different combinations of filters were tested using
forward selection. The best performing filter subset (collectively
termed "FP filters") consisted of a combination of 5 filters
(I+IV+V+VI+VIII) that achieved a combined rate of FP removal of
58.9%. In comparison, the combination of filters IV+VIII accounted
for most (54.4%) of the FPs eliminated, with incremental
improvements resulting from .ltoreq.5 additional filters.
Performance of these filters was evaluated on 3 sets of metaphase
images (Table 7), consisting of 2 HC image sets (HC-low and
HC-high, which were used during filter development) and an
independent low dose image set from CNL. On average, 55.+-.9.6% of
FPs were removed among all sets; individually the filters
eliminated 52% of FPs from the CNL set, which was comparable to the
HC sets (66% and 48% for low and high dose sets, respectively). All
TPs were retained in each of the sets after processing of FPs (i.e.
100% specificity).
Dose-response calibration curves for HC and CNL data were generated
in ADCI to investigate the effect of the filters on dose estimation
accuracy (FIG. 5). Dose accuracy was assessed by determining the
absolute error (absolute difference between dose estimate and true
physical dose). For comparison, the dose estimates of 6 test
samples (3 from HC, 3 from CNL) were compared which were either
unfiltered and in which combinatorial FP filters were applied
(Table 8). In samples that were manually curated by CNL, accuracy
was also improved >2-fold by applying the 5 combined FP filters
(average error decreased from 0.43 Gy to 0.18 Gy).
The dose accuracy in the HC samples was impacted by addition of
these filters (mean absolute error increased from 0.85 Gy to 1.03
Gy). One explanation was either the filters were removing many TPs
inadvertently, or FPs removed by the filters were offsetting
previously undetected DCs (false negatives) in the HC samples. All
objects eliminated with these filters in the 3 HC samples were
reviewed and classified as either TP or FP, and the FP specificity
across the samples was determined (Table 9). Similar to earlier
findings, the FP filters exhibited very high specificity for FPs
(97.7-100%), indicating that the filters retained high specificity
for TPs in the HC samples.
We hypothesized that the difference in image selection protocols
was responsible for the discrepancies seen in classification
performance and dose estimation accuracy between the two sources.
While CNL manually selected for images deemed suitable for DCA
analysis, image selection at HC was done with an automated
metaphase classifier that effectively removed only images lacking
metaphases (see Methods 1). Manual review of images in the HC and
CNL samples confirmed noticeable differences in image quality: In
concordance with findings from our previous study.sup.1, CNL data
contained more images with well-spread, minimally-overlapping
chromosomes, and fewer images with extreme SCS and chromosome
fragments (complete dissociation of sister chromatids). The HC data
contained a greater percentage of high-band-level (less condensed)
chromosomes, characteristic of prometaphase/early-metaphase cell
images. These chromosomes were the source of many unfiltered FPs,
due to the lack of a strong primary constriction at the
centromere.
A new set of HC calibration curves were then generated from
manually curated, selected images from calibration samples (FIG.
6). Images were excluded based on IAEA criteria.sup.17, along with
cells exhibiting long chromosomes in early prometaphase.sup.16
(Methods 5). (Table 10). Dose estimation accuracy of the HC samples
(INTC03S01, INTC03S08 and INTC03S10) was significantly improved by
enabling the 5 FP segmentation filters (mean unfiltered absolute
error was 0.37 Gy, and was 0.15 Gy with the filters; Table 10).
Therefore, application of FP filters to both CNL and curated HC
data led to >2-fold reduction in the mean absolute error of the
estimated dose (p=0.024, paired two tailed t-test).
Example 2. Application of Image Selection Models
Assessment of image selection was challenging, as no objective
standard exists. Cell selection by cytogenetic experts is based on
their knowledge of metaphase chromosome conformation, sensitivity,
and even individual preferences in interpreting images which are
sometimes inconsistent. Therefore, image selection methods were
evaluated through dose estimation of filtered test samples and
comparisons with known physical exposures. The images in all
calibration and test samples were processed by the same image
selection method. Dose estimates of test samples are calculated
using a curve fit to calibration samples. Dose estimation errors
indicate the accuracy of dicentric chromosome detection, and
consequently imply the effectiveness of image selection method.
To rank images with the combined z-score method, a weight vector
corresponding to each of the 6 filters comprising the total score
was first determined. Optimal weights were obtained by searching a
large number of possible values among the set of HC calibration
samples for those exhibiting smallest residuals when fit to the
curve. The potential weights were defined as integers ranging from
[1, 5]. This limited the search space and eased computational
complexity, but nevertheless ensured that diverse combinations of
weights were evaluated. In experiments, three optimal weight
vectors, namely [5, 2, 4, 3, 4, 1] [4, 3, 4, 5, 2, 1] and [1, 2, 1,
5, 1, 5], were used for dose estimation.
After images were assigned scores and sorted according to their
combined z-scores (or by the chromosome group bin method--see
below), the 250 top ranked images were subsequently selected to
determine dicentric aberration frequency for that sample. An
adequate number of top ranked images are selected to provide
sufficient images to generate a reproducible DC frequency for that
sample. The top ranked image set also has to effectively remove
poor quality images that could distort the DC frequency. IAEA has
recommended at least 100 DCs be detected for samples with physical
doses >1 Gy. In practice, laboratories score >250 images, but
often more. Considering the total number of images in a sample
ranges from 500 to 1500, we found that selecting the 250 top
scoring images gave satisfactory results. Upon integration of ADCI
with Metafer metaphase cell detection software, the time required
for the automated microscope system to capture a set of images in a
sample will be reduced by at least 50% and as much as 600%. FIG. 7
indicates that the DC frequency for the HC3 Gy calibration sample
stabilizes after at least 250 to 300 top images were included.
Similar results were obtained for other test and calibration
samples (not shown). DC frequencies can differ between image
selection methods because each method can select different images.
When the number of top ranked images significantly exceeds 300
images, differences between the specific image selection methods
are minimized as they share increasing numbers of selected images.
Unfiltered randomly sampled images from this sample tend to exhibit
higher overall DC frequencies due to increased numbers of FP
DCs.
The deviations of estimated doses of all of the HC and CNL test
samples, respectively, from physical doses, were determined for
various ADCI image selection models (Tables 10 and 11). For
comparison, the dose estimation results of unselected,
comprehensive sets of images for each sample are presented.
Deviations of <0.5 Gy from their calibrated physical dose are
acceptable for triage biodosimetry.sup.5,12. For the unfiltered HC
samples, the average absolute error is 0.8 Gy, with a single
sample, INTC03S01, fulfilling the triage criteria. The image
selection model that combines filters I-III and chromosome group
bin method produces the best result. Dose estimates for four
samples (INTC03S01, INTC03S08, INTC03S10 and INTC03S05) are
acceptable. The combined z-score method with the filter weights:
[1, 2, 1, 5, 1, 5] resulted in the least accurate estimates. Here,
the average error is .about.1 Gy, and only INTC03S05 had an
acceptable dose estimate. Of the five unfiltered CNL samples, only
INTC03S08 had an acceptable dose estimate. After applying image
selection models, a pan-filter set using all of the available
filters I-VI gave the most accurate results. The average absolute
error was .about.0.3 Gy, and 4 of 5 samples (INTC03S08, INTC03S04,
INTC03S05 and INTC03S07) exhibited doses in the acceptable
range.
Image selection rejects poor images and reduces FP DCs if
sufficient quantities of images remain to provide reliable DC
frequencies. Although >250 images were usually present after
scoring and ranking, application of image filters can result in
fewer remaining images for analysis. After applying the pan-filter
set, sample CNL-INTC03S08 consisted of 195 metaphase cells. After
applying the combined image selection model to the HC samples,
sample HC-INTC03S07 consisted of only 109 metaphase cells. This
sample was relatively lower quality than others in this set, and
the unfiltered set of metaphase images was smaller than the
recommended minimum (500 cells), consisting of 477 cells (Table
12).
Example 3. Automated Search for Optimized Image Selection
Models
The total number of potential image selection models is unlimited
and effective image selection model configurations differ between
laboratories due to their laboratory specific sample preparation
procedures. To alleviate the need to manually create and test sets
of image selection models, ADCI provides automated search
functionality to locate optimal models for given samples. The
automated search may take several hours to complete, depending on
the size of the search configuration space.
Automated searches of optimal image selection models involve two
steps: generation of a pool of possible image selection models and
evaluation of each model in the pool.
Generating Models
An image selection model consists of morphological filters and/or
image scoring. Each filter can either be enabled at a user
specified threshold value or be disabled altogether. Images can be
scored using the combined z-score method (contents of an image
selection model heading) or group bin method. The combined z-score
method requires a weight vector in which weights can be adjusted.
The number of selected top images after images are scored and
ranked is also adjustable. The automated search for optimal image
selection models has the capability to test all of these
configurations.
Image selection models are categorized in 3 groups:
(1) Filter-Only Models
A typical configuration for automated model generation in this
group is shown in the table below. A pool of selection models
containing all permutations of the filter thresholds listed in the
table are generated. Note each filter may also be in a disabled
state in addition to the values listed. Square brackets indicate a
pair of threshold values in the format: [lower bound, upper
bound].
TABLE-US-00001 Threshold Filters Filtering method values to test
Length-Width Exclude if length-width ratio 1.0, 1.5, 2.0 Ratio
z-score is > threshold Centromere Exclude if centromere density
1.0, 1.5, 2.0 Density z-score is > threshold Finite Exclude if
finite difference -1.0, -1.5, -2.0 Difference z-score is <
threshold Object Count Exclude if count is < lower [40, 60],
[40, bound or > upper bound 65] Segmented Exclude if count is
< lower [35, 50] Object Count bound or > upper bound
Classified Object Exclude if the ratio 0.6, 0.7 Ratio is <
threshold
(2) Combined z-Score Models without Filtering
Image selection models in this group use a weight vector to score
and rank images, then select a certain number of top images.
Typically, values 0, 1, 2, 3, 4, 5 are tested for weights and
numbers 250, 300, 400, 500 are tested for selecting top images.
(3) Filter and then Group Bin Models
In this group, a model will first apply filtering and then use the
group bin method to score and rank images. Configurations in group
1 and group 2 can be used to generate models in this group.
The total number of generated image selection models can be very
large. If the configurations shown above are used, 192,384 models
are generated.
Model Evaluation
Image selection models in the pool can be assessed using a set of
samples with known physical doses. An image selection model is
applied to all evaluation samples. Sample quality after image
selection is evaluated by one of the user-specified methods listed
below, which return a score indicating the effectiveness of the
model. Three evaluation methods can be selected:
1. (a) P-Value of Poisson Fits Each evaluation sample calculates a
p-value of its Poisson fit, determined by a user-specified SVM
sigma. P-values of all samples combine to a single score through
the use of Fisher's method. The score will be `nan`, not-a-number,
if any evaluation sample gives `nan` p-value, making the evaluation
invalid. When it happens, please try a larger sigma value or use
other evaluation methods.
2. (b) Curve Fitting Residual After users specify an SVM sigma
value, all evaluation samples are used to fit a calibration curve.
The squares of samples' fitting residuals on the curve are summed
to a single score.
3. (c) Leave-One-Out Dose Estimation Users first specify an SVM
sigma value. One sample in the evaluation set is used as a dose
estimation test sample. While the remaining are used for curve
calibration from which the dose estimation error is calculated for
the test sample. The process is repeated until every evaluation
sample has been used as test sample. The dose estimation error of
all these tests are squared and summed to form a single score.
Image selection models with the lowest scores are the optimal
models resulting from the search.
Implementation of the Optimized Image Selection Search Wizard.
Configure Image Selection Model Generation.
This is an optional configuration step. Default values are
prefilled in this dialog whether it is opened or not. To facilitate
automated. generation of image selection models, the software
stores multiple options for filter thresholds, weights of the
combined z-score, and number of selected top images. These values
can be adjusted by users. To open this dialog click "Settings" in
the menu bar at the top of the software window and select "Image
Selection Optimization Settings". The dialog to the right will be
displayed and configuration changes can be made within the
dialog.
Automated searches for optimal image selection model start from
"Image Selection Optimization" wizard. It can be opened from the
"Wizards" menu. A step by step guide to the wizard is provided
below.
Introduction.
Before proceeding to the next steps of the wizard, processed
samples must be present within the main GUI. The "curve fit
residual" evaluation method will require at least 3 samples, the
"leave-1-out dose estimation" method at least 4 samples.
Select Samples.
Sample selection in this wizard is the same as in the "Curve
Calibration" wizard. Selected samples will be used to evaluate
image selection models during optimization. Enter physical doses of
these samples if they are not auto-filled or if they were
auto-filled incorrectly.
Select an SVM Sigma Value.
Select an SVM sigma to use for optimization. It determines
dicentric chromosome distributions, which are used to calculate
p-values of Poisson fit, and dicentric chromosome frequencies,
which are used in the "curve fit residual" method and the
"leave-1-out dose estimation" method.
Configure Image Selection Model Search.
As described previously, possible image selection models are
logically categorized into 3 groups. In ADCI, users have the option
to include or exclude a group in the search of optimal models.
Place a checkmark beside groups that are intended to be searched.
Leave undesired groups unchecked. Image selection models in checked
groups will be generated according to configurations in "Image
Selection Optimization Settings".
Generally, it is desirable to select all parameter groups to
perform a full search. However, if lengthy compute time is
concerning or the search is only to be used for a quick test or
tutorial, the number of image selection models to be searched can
be reduced by leaving some of groups unselected.
Select and Evaluation Method.
Select one of the three evaluation methods to determine which
models exhibit the best accuracy for that method. Please recall
that "Curve Fit Residuals" requires at least 3 selected samples to
work correctly, and "Leave-1-out Dose Estimation Errors" needs at
least 4 selected samples.
Summary.
Ensure the previous selections are correct on the summary screen.
Note values entered on previous screens can be edited by clicking
the blue button on the top left of the wizard dialog. Click
"Finish" to complete the wizard and bring up the "Optimal Image
Selection Model Search" dialog.
Optimal Image Selection Model Dialog.
The automated search for optimal image selection models is
performed in this dialog. A summary, including model generation
configuration and evaluation method and the samples to be
evaluated, is shown in the top part of the screen. Users can verify
if the search parameters are correct before beginning the search,
determine the time for search, generate a report when complete or
abort the wizard.
Click the "Start" button to start the search. The progress will be
indicated by a progress bar. The entire search may take a few
minutes to a few hours, depending on the number of models being
searched, evaluation method ("leave-1-out dose estimation" method
will take longer time), and computer hardware. Users can abort the
search any time by clicking the "Abort" button.
When the search finishes, optimal image selection models will be
displayed in ascending order of evaluation score in the "Search
Result" panel. Models are named according to their automatically
assigned numbers during model generation. The evaluation score of
each model is displayed along with the model in the list. The list
shows 10 best models by default. Up to 50 best models can be
displayed by clicking the "More" button.
After selecting an image selection model in the list, its content
will be shown in the widget to the right of the list. It is the
same widget used for image selection models in "Metaphase Viewer",
"Curve Calibration" wizard and "Dose Estimation" wizard. Please
note that any modification made to the widget will not change the
actual model.
A selected image selection model can be saved by clicking the
"Save" button. Its evaluation performance on each sample can be
viewed by clicking the "View" button and specifying an evaluation
sample. If the evaluation method is "p-values of Poisson fit", the
plot panel will show the sample's Poisson fit. Similarly, the plot
panel will show calibration curve and dose estimation for the
methods "Curve fit residual" and "Leave-1-out dose estimation
errors", respectively.
After an optimal image selection model has been saved, it will
appear alongside other saved image selection models and preset
image selection models wherever image selection models are listed
(FIG. 9). The metaphase image viewer can be used to view the
content of a saved selection model.
A request to prepare a "Report" displaying the optimal image
selection model parameters, and the associated calibration curve
and dose estimates for samples evaluated with the selected method
has been implemented in the final screen of the wizard.
Time Required to Perform Optimized Image Selection Model Search
The time required to perform a search depends on the number of
models being generated/examined. The number of models examined is
determined by the selected image selection model categories to
search (image filtering, combined z score, group bin) as well as
settings specified in "Settings".fwdarw."Image Selection
Optimization Settings". If the search space is large (>50,000
parameter combinations), ADCI may require hours to finish the
optimization.
Search runtime is proportional to the number of samples used for
evaluation. The particular evaluation method selected also
influences the search execution time. The `p-value of Poisson` and
`curve fit residual` methods take approximately the same amount of
time. The `Leave-one-out` method requires longer to complete, and
is proportional to the number of samples being evaluated. For
example, to finish a search using 10 evaluating samples,
`leave-one-out` will take approximately 10-times longer than either
`curve fit residual` or `p-value of Poisson`. These methods often
produce different optimal models and it is not possible to predict
which method will produce the best performing model.
Example 4. Sample Quality Assessment after Image Selection
To evaluate whether the image selection models improved sample
quality, a Chi squared goodness of fit test was performed on the
observed DC/cell vs. Poisson distributions for the CNL and HC
samples, both prior to and after automated and manual image
selection (Table 12). Manual image selection for CNL samples was
performed by CNL during sample preparation, while image selection
for HC samples was performed on unselected datasets (see Methods 5;
samples HC-INTC03S01, HC-INTC03S08, HC-INTC03S10 were analyzed,
despite <500 images being available). For each laboratory, the
best performing image selection models were used for FP and image
level filtering (Tables 10 and 11). Image selection with filters
I-III and chromosome group bin method was applied to HC sample
data, whereas filters I-VI were applied to the CNL data. At the 1%
significance level (i.e. Poisson goodness-of-fit, p.ltoreq.0.01),
86% (19 of 22) of unfiltered samples are significantly differed
from the Poisson distribution, and 76% (13 of 17) of manually--and
77% (17 of 22) of automatically-selected samples did not differ;
manually curated and uncurated sample groups also significantly
differed from each other (p=0.0021; one-tailed Wilcoxon Signed-Rank
Test, .alpha.=0.05, n=17). Therefore, the Poisson goodness of fit
measures changes in overall sample quality from image model
selection. While the Poisson score is improved for all of the
automatically selected datasets, the lowest quality samples (CNL1
Gy, CNL05 Gy, CNL-INTC03S01, HC-INTC03S05, HC-INTC03S07) were still
rejected as Poisson-distributed after automated filtering.
LITERATURE REFERENCES
Blakely, W. F., Salter, C. A., and Prasanna, P. G. (2005)
Early-response biological dosimetry--recommended countermeasure
enhancements for mass-casualty radiological incidents and
terrorism, Health physics 89, 494-504. Wilkins R C, Romm H, Kao T
C, Awa A A, Yoshida M A, Livingston G K, Jenkins M S, Oestreicher
U, Pellmar T C, Prasanna P G (2008) Interlaboratory comparison of
the dicentric chromosome assay for radiation biodosimetry in mass
casualty events. Radiation Research 169, 551-560. Bauchinger M.
Cytogenetic effects in human lymphocytes as a dosimetry system. In:
Eisert W S, Mendelsohn M L eds. Biological dosimetry: Cytometric
approaches to mammalian systems. Berlin: Springer-Verlag; 15-24
(1984). Lloyd D C, Edwards A A, Prosser J S. Chromosome aberrations
induced in human lymphocytes by in vitro acute X and Gamma
radiation. Rad. Prot. Biodosimetry. 15:83-88 (1986). International
Atomic Energy Agency (IAEA). Cytogenetic analysis for radiation
dose assessment. Technical Report Series No. 405, Vienna (2001).
International Atomic Energy Agency (IAEA). Cytogenetic Dosimetry:
Applications in Preparedness for and Response to Radiation
Emergencies, Vienna (2011). Schunck C., Johannes T, Varga D, Lorch
T and Plesch A, New developments in automated cytogenetic imaging:
unattended scoring of dicentric chromosomes, micronuclei, single
cell gel electrophoresis, and fluorescence signals. Cytogenet.
Genome Res. 104, 383-389 (2004). Ainsbury, E. A., Livingston, G.
K., Abbott, M. G., Moquet, J. E., Hone, P. A., Jenkins, M. S.,
Christensen, D. M., Lloyd, D. C. and Rothkamm, K. Interlaboratory
variation in scoring dicentric chromosomes in a case of
partial-body x-ray exposure: implications for biodosimetry
networking and cytogenetic `triage mode` scoring. Radiat. Res. 172:
746-752 (2009). Lloyd D C, Edwards A A, Moquet J E,
Guerrero-Carbajal Y C. The role of cytogenetics in early triage of
radiation casualties. Appl. Radiat. Isot. 52:1107-1112 (2000).
Flegal, F N., Devantier, Y., McNamee, J P., Wilkins, R C. Quickscan
dicentric chromosome analysis for radiation biodosimetry. Health
Phys. 98: 276-281 (2010). Vaurijoux, A., Gruel, G., Pouzoulet, F.,
Gregoire, E., Martin, C., Roch-Lef{grave over (w)}vre, S., Voisin,
P., Voisin, P. and Roy, L., Strategy for population triage based on
dicentric analysis. Radiat Res 171:541-548 (2009). Vaurijoux A,
Gruel G, Gregoire E, Roch-Lefevre S, Voisin P, Martin C, Voisin P,
Roy L, and Barquinero J-F. Automatic Dicentric Scoring a Real
Option to Be Used in Biological Dosimetry. Radiation Emergency
Medicine. 4:16-21 (2015). Gruel G., Gregoire E, Lecas S, Martin C,
Roch-Lefevre S, Vaurijoux A, Voisin P, Voisin P, and Barquinero,
J-F. Biological Dosimetry by Automated Dicentric Scoring in a
Simulated Emergency. Radiation Res. 179: 557-569 (2013) Zhou, W.,
Bovik, A. C., Sheikh, H. R., and Simoncelli, E. P. (2004) Image
quality assessment: from error visibility to structural similarity,
IEEE Transactions on Image Processing 13, 600-612. Nill, N. B., and
Bouzas, B. (1992) Objective image quality measure derived from
digital image power spectra, OPTICE 31, 813-825. Narwaria, M., and
Lin, W. (2010) Objective Image Quality Assessment Based on Support
Vector Regression, IEEE Transactions on Neural Networks 21,
515-519. Li, Y., Knoll, J. H. M., Wilkins, R. C., Flegal, F. N.,
and Rogan, P. K. (2016) Automated discrimination of dicentric and
monocentric chromosomes by machine learning-based image processing,
Microscopy Research and Technique 79, 383-402. Arachchige, A. S.,
Samarabandu, J., Knoll, J., Khan, W., and Rogan, P. (2010) An image
processing algorithm for accurate extraction of the centerline from
human metaphase chromosomes, In Image Processing (ICIP), 2010 17th
IEEE International Conference on, pp 3613-3616. Arachchige, A. S.,
Samarabandu, J., Rogan, P. K., and Knoll, J. H. M. (2012) Intensity
integrated Laplacian algorithm for human metaphase chromosome
centromere detection, In Electrical & Computer Engineering
(CCECE), 2012 25th IEEE Canadian Conference on, pp 1-4. Arachchige,
A. S., Samarabandu, J., Knoll, J. H., and Rogan, P. K. (2013)
Intensity integrated Laplacian-based thickness measurement for
detecting human metaphase chromosome centromere location, IEEE
transactions on bio-medical engineering 60, 2005-2013. Subasinghe,
A., Samarabandu, J., Li, Y., Wilkins, R., Flegal, F., Knoll, J. H.
M., and Rogan, P. K. (2016) Centromere Detection of Human Metaphase
Chromosome Images using a Candidate Based Method, F1000Research 5,
1565, 2016. Rogan, P. K., Li, Y., Wilkins, R. C., Flegal, F. N.,
and Knoll, J. H. M. (2016) Radiation Dose Estimation by Automated
Cytogenetic Biodosimetry, Radiation Protection Dosimetry 172,
207-217. Rieder, C. L., and Palazzo, R. E. (1992) Colcemid and the
mitotic cycle, Journal of cell science 102 (Pt 3), 387-392.
Sethakulvichai, W., Manitpornsut, S., Wiboonrat, M., Lilakiatsakun,
W., Assawamakin, A., and Tongsima, S. (2012) Estimation of band
level resolutions of human chromosome images, In Computer Science
and Software Engineering (JCSSE), 2012 International Joint
Conference on, pp 276-282. Carothers, A., and Piper, J. (1994)
Computer-aided classification of human chromosomes: a review,
Statistics and Computing 4, 161-171. International Standing
Committee on Human Cytogenetic Nomenclature, Shaffer, L. G.,
McGowan-Jordan, J., and Schmid, M. (2013) ISCN 2013: An
International System for Human Cytogenetic Nomenclature (2013),
Karger.
TABLES
TABLE-US-00002 TABLE 1 Metaphase image sets used in development and
validation of DC filters. HC-mixed* Dataset Name HC-low HC-high
CNL-low Health Health Canadian Nuclear Lab source Canada Canada
Laboratories Radiation dose (Gy) 1 3-4 1 No. of images 198 216 256
No. of chromosomes** 8041 8697 10583 No. of TPs 20 163 14 No. of
FPs 97 61 82 *HC-mixed refers to a combined set of all images from
both the HC-low + HC-high datasets **Defined as number of valid
segmented objects defined by ADCI.
TABLE-US-00003 TABLE 2 Metaphase image samples used in construction
of dose calibration curves. Sample Physical No. of No. of (HC or
CNL) xGy dose images, HC images, CNL 0 Gy 0 Gy 731 798 0.5 Gy 0.5
Gy 586 1532 1 Gy 1 Gy 1566 841 2 Gy 2 Gy 1147 996 3 Gy 3 Gy 1212
1188 4 Gy 4 Gy 909 1635
TABLE-US-00004 TABLE 3 Metaphase image samples used in evaluation
of dose assessment performance. Physical No. of No. of images, Dose
images, HC.sup.1 CNL.sup.2 Sample name (Gy) preparation preparation
INTC03S01 3.1 540 500 INTC03S08 2.3 637 500 INTC03S10 1.4 708 n/a
INTC03S04 1.8 996 957 INTC03S05 2.8 1136 1527 INTC03S07 3.4 477 735
.sup.1HC: Health Canada. .sup.2CNL: Canadian Nuclear Laboratory.
n/a: sample data were not available.
TABLE-US-00005 TABLE 4 Comparison of FP subclass targeting between
proposed DC filters. No. FP DCs removed filtering by morphological
subclass* SCS Fragment Overlap Noise Debris ML DC filter
designation** (n = 51) (n = 10) (n = 17) (n = 5) (n = 4) (n = 11)
I: Area 19 6 0 0 3 1 II: Mean width 14 6 0 0 2 0 III: Median width
12 5 0 0 3 0 IV: Max width 23 8 0 0 3 0 V: Centromere width 8 3 0 0
3 0 VI: Oblongness 31 1 2 0 0 0 VII: Contour symmetry 11 0 0 0 0 0
VIII: Intercandidate 43 2 3 1 1 2 contour symmetry *See Methods 2
for description of each subclass. Calculated from HC-low image set
in Table 1. **See Methods 3.1 for description of each filter.
TABLE-US-00006 TABLE 5 Comparison of FP discrimination ability
between proposed DC filters. 2-sample FP K-S, TPs/FPs, removed DC
filter designation** p-value* (%)* I: Area 2.2 .times. 10.sup.-18
22.2 II: Mean width 9.2 .times. 10.sup.-10 16.5 III: Median width
3.3 .times. 10.sup.-9 14.6 IV: Max width 3.3 .times. 10.sup.-8 27.8
V: Centromere width 8.8 .times. 10.sup.-3 13.9 VI: Oblongness 1.1
.times. 10.sup.-24 27.2 VII: Contour symmetry 1.2 .times. 10.sup.-8
10.1 VIII: Intercandidate contour symmetry 4.0 .times. 10.sup.-30
44.9 *Calculated from HC-mixed image set from Table 1. **See
Methods 3.1 for description of each filter.
TABLE-US-00007 TABLE 6 Forward selection results by combining
subsets of DC filters. FP removed DC filter subset** (%)* 1-filter:
VIII 44.9 2-filters: VIII + IV 54.4 3-filters: VIII + IV + V 56.3
4-filters: VIII + IV + V + VI 58.2 5-filters: VIII + IV + V + VI +
I 58.9 *Calculated from HC-mixed image dataset from Table 1. **See
Methods 3.1 for description of filters.
TABLE-US-00008 TABLE 7 Performance evaluation of FP filters* on
development and validation image datasets. Image No. of TP No. of
FP FP removed set** DCs removed DCs removed (%) HC-low 0 64 66
HC-high 0 29 48 CNL-low 0 43 52 *FP filters refer to the subset of
filters I + IV + V + VI + VIII (see Methods 3.1). **See Table 1 for
sample details.
TABLE-US-00009 TABLE 8 Dose estimation of test samples, with and
without FP filters* enabled. HC samples** CNL samples** Physical
dose (Gy) 3.1 2.3 1.4 1.8 2.8 3.4 Estimate, unfiltered (Gy) 3.90
1.65 0.30 1.35 2.40 2.95 Estimate, FP filters (Gy) 2.45 1.25 0.00
2.1 2.75 3.55 *FP filters refer to the subset of filters I + IV + V
+ VI + VIII (see Methods 3.1). Calibration curve image data was not
curated or filtered. HC samples were unselected (INTC03S01,
INTC03S08, and INTC03S10). The CNL samples were previously manually
curated (INTC03S04 [n = 448], INTC03S05 [n = 500], and INTC03S07 [n
= 385). **See Table 3 for sample details.
TABLE-US-00010 TABLE 9 Specificity of FP filters* in HC test
samples. Total no. of No. No. Specificity chromosomes of TPs of FPs
for Image sample** removed removed removed FPs INTC03S01 193 0 193
100% INTC03S08 133 3 130 97.7% INTC03S10 143 2 141 98.6% *FP
filters refer to the subset of filters I + IV + V + VI + VIII (see
Methods 3.1). **See Table 3 for sample details.
TABLE-US-00011 TABLE 10 Dose estimates and deviations from physical
dose for HC test samples after applying image selection models.
Image selection INTC03S01 INTC03S08 INTC03S10 INTC03S04 INTC03S05
INTC03S07 model 3.1.sup. 2.3 1.4 1.8 2.8 3.4 All images 2.65, -0.45
1.4, -0.9 0.15, -1.25 3.05, +1.25 2.2, -0.6 3.95, +0.55 Combined z
2.75, -0.35 2, -0.3 1.35, -0.05 2.85, +1.05 2.55, -0.25 4, +0.6
score, weight [5, 2, 4, 3, 4, 1], top 250 Combined z 2.85, -0.25 2,
-0.3 1.25, -0.15 2.7, +0.9 2.4, -0.4 4, +0.6 score, weight [4, 3,
4, 5, 2, 1], top 250 Combined z 1.6, -1.5 1.4, -0.9 0.5, -0.9 3.6,
+1.8 2.6, -0.2 4, +0.6 score, weight [1, 2, 1, 5, 1, 5], top 250
Chromosome 2.55, -0.55 2.25, -0.05 1.1, -0.3 2.45, +0.65 2.75,
-0.05 2.15, -1.25 group bin method, top 250 Filters I-VI 2.05,
-1.05 1, -1.3 0.35, -0.95 1.55, -0.25 2.05, -0.75 1.2, -2.2 Filters
I-III & 2.8, -0.3 1.95, -0.35 1, -0.4 2.35, +0.55 2.8, +0.0
2.25, -1.15 chromosome group bin method, top 250 Manual image 2.85,
-0.25 2.4, +0.1 1.25, -0.15 n/a n/a n/a curation n/a: manual
selection result not available. .sup. Sample identifier, physical
dose (Gy). FP filters were enabled.
TABLE-US-00012 TABLE 11 Dose estimates and deviations from physical
dose for CNL test samples after applying image selection models.
INTC03S01 INTC03S08 INTC03S04 INTC03S05 INTC03S07 Image selection
model 3.1{circumflex over ( )} 2.3 1.8 2.8 3.4 All images 4, +0.9
2.6, +0.3 2.45, +0.65 3.6, +0.8 4, +0.6 Combined z score, weight
[5, 3.95, +0.85 2.8, +0.5 2, +0.2 3, +0.2 3.55, +0.15 2, 4, 3, 4,
1], top 250 Combined z score, weight [4, 4, +0.9 2.7, +0.4 1.65,
-0.15 3.05, +0.25 3.95, +0.55 3, 4, 5, 2, 1], top 250 Combined z
score, weight [1, 3.6, +0.5 2.4, +0.1 0.65, -1.15 2.35, -0.45 3.05,
-0.35 2, 1, 5, 1, 5], top 250 Chromosome group bin 4, +0.9 2.8,
+0.5 1.75, -0.05 2.5, -0.3 4, +0.6 method, top 250 Filters I-VI
3.75, +0.65 2.8, +0.5 1.9, +0.1 3.05, +0.25 3.4, +0.0 Filters I-III
& chromosome 4, +0.9 2.75, +0.45 1.65, -0.15 2.25, -0.55 3.95,
+0.55 group bin method, top 250 Manual image curation n/a n/a 2.1,
+0.3 2.75, -0.05 3.55, +0.15 n/a: manual selection result not
available. {circumflex over ( )}Sample identifier, physical dose
(Gy). FP filters were enabled.
TABLE-US-00013 TABLE 12 Goodness of fit Poisson scores* of
unfiltered, manually- and ADCI-filtered image sets for calibration
and test samples. Auto- mated image selection: mor- Auto- phology
mated filters image and selection: chromo- mor- Manual some phology
image group bin filters Sample All images selection
method{circumflex over ( )} only.sup.# HC0Gy 1.333e-15 2.240e-01
NaN n/a.sup.@ HC05Gy 1.232e-01 unavailable 3.637e-01 n/a HC1Gy
1.669e-18 9.996e-01 1.049e-01 n/a HC2Gy 4.019e-64 2.072e-01
7.618e-04 n/a HC3Gy 2.873e-02 4.642e-01 6.112e-01 n/a HC4Gy
2.596e-04 2.215e-01 3.127e-01 n/a HC-INTC03S01 <2.225e-308.sup.+
9.052e-01 1.170e-01 n/a HC-INTC03S08 1.236e-01 4.573e-01 8.153e-01
n/a HC-INTC03S10 8.873e-01 3.895e-01 2.113e-01 n/a HC-INTC03S04
0.000e+00 unavailable 2.931e-02 n/a HC-INTC03S05 1.103e-06
unavailable 3.544e-03 n/a HC-INTC03S07 <2.225e-308 unavailable
1.996e-04 n/a CNL0Gy 5.174e-03 1.254e-01 n/a 3.071e-01 CNL05Gy
1.656e-157 1.236e-01 n/a 5.955e-32 CNL1Gy 9.801e-30 1.496e-03 n/a
1.597e-06 CNL2Gy 2.340e-147 <2.225e-308 n/a 4.488e-02 CNL3Gy
8.489e-07 6.820e-03 n/a 9.914e-01 CNL4Gy 5.151e-22 3.303e-02 n/a
1.826e-01 CNL-INTC03S04 1.728e-60 1.933e-02 n/a 5.446e-02
CNL-INTC03S05 2.743e-09 5.243e-02 n/a 3.253e-01 CNL-INTC03S07
6.671e-10 4.248e-05 n/a 4.725e-01 CNL-INTC03S01 <2.225e-308
unavailable n/a 7.627e-11 CNL-INTC03S08 5.253e-16 unavailable n/a
7.768e-01 *Poisson score is the p-value of chi-square goodness of
fit (without merging bins) of observed distnbution of DCs/cell vs.
Poisson distribution determined from average DC frequency.
Filtering parameters chosen for each laboratory exhibit dose
estimates that are closest to the physical dose: {circumflex over (
)}HC image sets were filtered with morphological filters I-III and
by chromosome group bin score; .sup.#CNL image sets were filtered
with morphological filters I-VI. .sup.@n/a: not applicable, the
other demonstrated image selection method has better dose
estimation result. .sup.+Minimum positive floating value in Windows
operating system. NaN: P-value could not be determined due to
insufficient degrees of freedom. Unavailable: manual image
selection was not performed.
* * * * *